public-data

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

(★ 9)

Curated GenAI patterns, prompts, and agent architectures extracted from open-source projects

public-data 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# public-data

Curated patterns, prompts, architectural ideas, and repo reviews extracted from publicly available AI projects.

## Publishing Policy

**What we share:**
- Summaries and useful parts of any publicly available code repository
- Full write-ups in `reviews/` for every repo reviewed
- Extracted patterns, prompts, and tools in their respective directories

**Attribution:**
- Every file includes source repo, author, and license
- Files with no license are marked clearly: "no license specified — educational/personal use only"
- We respect that authors chose to share publicly. We honor that by sharing our analysis publicly too, with attribution.

**License handling:**
- Apache-2.0, MIT, BSD, CC-BY → extract freely, include attribution
- AGPL/GPL → summarize and document patterns; do not embed in proprietary code
- No license → note explicitly; treat as educational/personal use only; do not redistribute code itself
- Always attribute: source repo, author, and license in each file

---

## Structure

- `reviews/` — Full write-ups of every repo reviewed (published with attribution)
- `prompts/` — System prompts, prompt templates, personality definitions
- `agents/` — Agent definitions, orchestration configs, loop architectures
- `patterns/` — Architectural patterns, best practices, reusable designs
- `tools/` — Useful scripts and utilities
- `notes/` — Research notes and analysis

---

## Reviews

Full write-ups of repos reviewed, with architecture summaries, key patterns, code examples, and relevance notes.

Repository tracking:
- [research-channel-repositories.md](reviews/research-channel-repositories.md) — GitHub repositories shared in the #research Discord channel, with review status and last review date.

| File | Source | License | Rating | Description |
|------|--------|---------|--------|-------------|
| [diagram-design.md](reviews/diagram-design.md) | [cathrynlavery/diagram-design](https://github.com/cathrynlavery/diagram-design) | MIT | ✅ Deploy candidate | Agent Skill/plugin for branded editorial technical diagrams with 27 visual types, semantic behavior patterns, draw.io and Mermaid redraw flows, static-first HTML/SVG output, accessibility checks, constrained motion, and broad validation gates |
| [deepseek-harness.md](reviews/deepseek-harness.md) | [deepseek-ai/deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) | MIT | ✅ Deploy candidate | DeepSeek's developer-preview agent harness with Cordis plugin composition, web/headless profiles, ACP automation, Python runtime, event-sourced sessions, guarded tool execution, sandbox/approval seams, generated docs, and broad test coverage |
| [logseq.md](reviews/logseq.md) | [logseq/logseq](https://github.com/logseq/logseq) | AGPL-3.0 | ✅ Deploy candidate | Privacy-first local knowledge graph and outliner with Clojure/ClojureScript, Electron, DataScript, SQLite/DB graphs, Markdown/Org, plugins, sync/RTC, mobile apps, CLI, and MCP surfaces; mature but AGPL/DB-beta/dependency-audit caveats |
| [jellyglance.md](reviews/jellyglance.md) | [Nerdy-Technician/JellyGlance](https://github.com/Nerdy-Technician/JellyGlance) | MIT/GPL metadata conflict | ⚠️ Interesting | Self-hosted Jellyfin operations dashboard with Express, React/Vite, PostgreSQL, live sessions, users, stats, Arr/Seerr/download integrations, webhooks, backups, and tasks; promising but needs auth, secret, license, and test-posture hardening |
| [linkbreeze.md](reviews/linkbreeze.md) | [Manak-hash/LinkBreeze](https://github.com/Manak-hash/LinkBreeze) | MIT | ⚠️ Interesting | Self-hosted Linktree alternative built with Next.js, React, Drizzle, SQLite, Docker, multi-page profiles, migration import, QR codes, themes, email capture, and privacy-first analytics; promising but needs production-secret and trusted-snippet hardening |
| [napkin.md](reviews/napkin.md) | [Michaelliv/napkin](https://github.com/Michaelliv/napkin) | MIT | ✅ Deploy candidate | Local-first TypeScript/Bun CLI and SDK for agent-operable Markdown/Obsidian vaults, with progressive disclosure via `NAPKIN.md`, overview, search, and read; broad notes/tasks/links/Bases/Canvas coverage, bundled distill/tend skills, passing tests, and packaging/audit caveats |
| [cloudflare-computer.md](reviews/cloudflare-computer.md) | [cloudflare/computer](https://github.com/cloudflare/computer) | MIT | ⚠️ Interesting | Cloudflare preview package for persistent agent workspaces with Durable Object SQLite filesystem, pluggable Worker shell/Worker JavaScript/container execution backends, AI SDK tools, git/assets/artifacts surfaces, and preview/API-stability caveats |
| [meat.md](reviews/meat.md) | [boldsoftware/meat](https://github.com/boldsoftware/meat) | Apache-2.0 | ✅ Deploy candidate | Go CLI that turns raw code diffs into source-derived reading diffs with OpenAI/Anthropic/exe.dev model support, read-only repo context tools, chunked large-diff handling, cache keys tied to prompt/compiler protocol, and a strong LLM edit-plan compiler boundary |
| [optimal-hypothesis-weakest-not-shortest.md](reviews/optimal-hypothesis-weakest-not-shortest.md) | [The Optimal Choice of Hypothesis Is the Weakest, Not the Shortest](https://arxiv.org/abs/2301.12987v4) | CC BY-NC-ND 4.0 | 📚 Good reference | AGI 2023 paper arguing that, under a finite enactive-cognition task formalism with uniform task distribution, the best valid hypothesis is the weakest one by extension size rather than the shortest one by description length; useful challenge to compression-as-intelligence, with toy binary-arithmetic evidence and strong assumption caveats |
| [inspect-ai.md](reviews/inspect-ai.md) | [UKGovernmentBEIS/inspect_ai](https://github.com/UKGovernmentBEIS/inspect_ai) / [Inspect docs](https://inspect.aisi.org.uk/) | MIT | ✅ Deploy candidate | UK AI Security Institute framework for LLM and agent evaluations with Python tasks, datasets, solvers, scorers, provider adapters, tool execution, sandboxes, approval policies, log viewer, dataframe/scanner analysis, 200+ evals, and a local detached-run control channel |
| [ori-eval.md](reviews/ori-eval.md) | [OpenRouter Ori Eval](https://openrouter.ai/ori/eval) / [OpenRouterLabs/ori-releases](https://github.com/OpenRouterLabs/ori-releases) | Apache-2.0 | ✅ Deploy candidate | OpenRouter CLI workflow for model and agent evals using `*.eval.ts`, Bun, generated `ori/eval` SDK, tool/completion/judge assertions, candidate-vs-judge cost reporting, scratch workspaces, OpenRouter catalog routing, and CI-friendly reports/history |
| [time-to-first-token.md](reviews/time-to-first-token.md) | [patchy631/time-to-first-token](https://github.com/patchy631/time-to-first-token) | Apache-2.0 repo; README content CC BY 4.0 | 📚 Study | Ten-week practical LLM inference-serving roadmap centered on one OpenAI-compatible service, with vLLM/SGLang, Prometheus/Grafana, GuideLLM/genai-perf, quantization, speculative decoding, disaggregation, queue-depth autoscaling, cost routing, and benchmark hygiene |
| [qwen38-mtp.md](reviews/qwen38-mtp.md) | [sudoingX/qwen38-mtp](https://github.com/sudoingX/qwen38-mtp) | Apache-2.0 | 🔧 Harvest | Minimal llama.cpp/Qwen3.8 MTP recipe showing one draft-speculative-decoding flag set, q4 KV cache config, paired streaming probe, and early 24GB-GPU/community measurements for roughly one-third faster decode |
| [context-governor.md](reviews/context-governor.md) | [gbgh1/context-governor](https://github.com/gbgh1/context-governor) | MIT | ⚠️ Interesting | Local OpenAI-compatible proxy plus MCP server for keeping long agent sessions bounded by storing bulky context off-wire, forwarding stable handle stubs, and rehydrating exact memory under budget; strong proxy core but current MCP dependency break blocks clean deployment |
| [hermes-bot-mode.md](reviews/hermes-bot-mode.md) | [NousResearch/Hermes-Bot-Mode](https://github.com/NousResearch/Hermes-Bot-Mode) | MIT | ✅ Deploy candidate | Hermes desktop plugin that turns existing profiles into named bots with canonical chats, avatars, profile editing, scoped routines, and bot-to-bot handoff protocol; strong pattern with recent-Hermes runtime coupling |
| [morpho.md](reviews/morpho.md) | [paradigms-of-intelligence/morpho](https://github.com/paradigms-of-intelligence/morpho) | Apache-2.0 | 📚 Study | Browser-native research prototype for MorphoHDL, a recursive structural HDL/graph-rewrite system with width-agnostic buses, fallback base cases, SoA JavaScript compiler, interactive circuit-growth visualizations, Python reference tests, and prototype/no-CI/no-package caveats |
| [open-kritt.md](reviews/open-kritt.md) | [Kritt-ai/open-kritt](https://github.com/Kritt-ai/open-kritt) | AGPL-3.0-only | ⚠️ Interesting | Self-hosted AI security-research platform that orchestrates Codex/Claude/OpenRouter agents over repos with workflow steps, structured findings, dedupe, post-scripts, severity rankers, and an unusually candid threat model; useful authorized-lab tool with privileged Docker/root-runner/no-auth caveats |
| [langfuse.md](reviews/langfuse.md) | [langfuse/langfuse](https://github.com/langfuse/langfuse) | MIT OSS tree; separate enterprise directories | ✅ Deploy candidate | Production-shaped open-source LLM engineering platform for traces, prompts, evals, datasets, playgrounds, and dashboards, with Next.js/TypeScript, Postgres, ClickHouse, Redis/BullMQ, object storage, serious CI/security posture, and multi-service self-hosting caveats |
| [gl4ss.md](reviews/gl4ss.md) | [elder-plinius/GL4SS](https://github.com/elder-plinius/GL4SS) | AGPL-3.0-or-later | ⚠️ Interesting | Browser-only spatiotemporal image/video generator with BYOK OpenRouter flow, CSP-bounded key handling, origin-scoped authorization, IndexedDB frame archive, demand/prefetch queues, WebGL/map/globe UI, and generated-history/AGPL/dev-maintenance caveats |
| [scroll-world.md](reviews/scroll-world.md) | [oso95/scroll-world](https://github.com/oso95/scroll-world) | MIT | ✅ Deploy candidate | Agent skill for cinematic scroll-scrubbed brand/world landing pages, with budget-gated media generation, frame-locked video seams, native mobile chain guidance, vanilla JS scrub engine, and paid-provider/trusted-config caveats |
| [scopey.md](reviews/scopey.md) | [ArchAstro/scopey](https://github.com/ArchAstro/scopey) | MIT | ✅ Deploy candidate | Rust hook-side scope monitor for Claude Code, Codex, Grok, Pi, and OpenCode sessions, with current-scope extraction, structured tool journaling, background judging, delayed correction injection, Herdr/OS notifications, Homebrew releases, and sensitive-log/pre-1.0 hook caveats |
| [dictionary-of-ai-coding.md](reviews/dictionary-of-ai-coding.md) | [mattpocock/dictionary-of-ai-coding](https://github.com/mattpocock/dictionary-of-ai-coding) | **no license** | 📚 Study | Plain-English AI coding glossary generated from 69 Markdown term entries, with strong model/harness/context/tool vocabulary, agent-editing rules, README freshness CI, and no-license/stale-generated-README caveats |
| [sageroute.md](reviews/sageroute.md) | [codejunkie99/sageroute](https://github.com/codejunkie99/sageroute) | MIT | ⚠️ Interesting | Trajectory-aware model router that exposes OpenAI/Anthropic-compatible proxy routes, starts sessions on a cheap model, derives stuck/progress signals from tool-call evidence, asks Levanto Sage at checkpoints, and applies a guarded cheap-to-strong ladder; strong pattern, but early ops posture |
| [openscience.md](reviews/openscience.md) | [synthetic-sciences/openscience](https://github.com/synthetic-sciences/openscience) | Apache-2.0 | ✅ Deploy candidate | Local AI workbench for scientific research with Bun/TypeScript agent runtime, SolidJS workspace, 42 scientific connectors, `science_search`/`science_fetch`, 294 bundled skills, MCP/ACP/plugin surfaces, strong release automation, and unsandboxed/dependency-audit caveats |
| [wandr.md](reviews/wandr.md) | [perplexityai/wandr](https://github.com/perplexityai/wandr) | Apache-2.0 | ✅ Deploy candidate | Harbor benchmark for wide/deep research agents with 501 generated tasks, required JSONL outputs, provider-neutral Relay adapters, fetch/triage/canon/dedup/judge scoring, and explicit cost/network caveats |
| [moonep.md](reviews/moonep.md) | [MoonshotAI/MoonEP](https://github.com/MoonshotAI/MoonEP) | MIT | 📚 Study | Expert-parallel MoE communication library that keeps per-rank token loads exactly balanced with dynamic redundant experts, static `S x K` buffers, zero-copy dispatch/combine views, online CUDA planning, NVLink/VMM symmetric memory, and multi-GPU test/benchmark caveats |
| [deepseek-v4-flash-dspark-dgx-spark.md](reviews/deepseek-v4-flash-dspark-dgx-spark.md) | [MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark](https://github.com/MiaAI-Lab/DeepSeek-v4-Flash-DSpark-2x-DGX-Spark) | MIT repo scripts/docs; upstream component terms vary | ⚠️ Interesting | Two-node DGX Spark vLLM/DeepSeek V4 Flash 0731 runbook with 1M-token NVFP4 KV profile, MTP-5 defaults, 0731 encoding/tool-call notes, env-lane split, benchmark script/raw results, regular CUDA graph evidence, and privileged-cluster caveats |
| [foundry.md](reviews/foundry.md) | [simoncorry/foundry](https://github.com/simoncorry/foundry) | MIT | ✅ Deploy candidate | Plain-markdown agent process for durable software work, with staged planning/framing/critique/build/test/wrap-up commands, repo-local plans/sessions/wiki, generated Cursor/Claude/Codex command shapes, zero dependencies, and passing local checks |
| [gigatoken.md](reviews/gigatoken.md) | [marcelroed/gigatoken](https://github.com/marcelroed/gigatoken) | MIT | ✅ Deploy candidate | Rust/Python tokenizer engine for GB/s language-model tokenization, with SIMD pretokenizers, repeated-pretoken caches, native file/JSONL/Parquet inputs, Hugging Face/tiktoken compatibility wrappers, and broad parity tests |
| [symphony.md](reviews/symphony.md) | [openai/symphony](https://github.com/openai/symphony) | Apache-2.0 | ✅ Deploy candidate | OpenAI's spec and Elixir/OTP reference implementation for issue-tracker-driven autonomous coding-agent runs, with repo-owned workflow contracts, Codex app-server sessions, multi-tracker adapters, host-side tracker tools, dashboard/API observability, CI, and release artifacts |
| [claudecodepsymphony.md](reviews/claudecodepsymphony.md) | [gherghett/ClaudeCodePSymphony](https://github.com/gherghett/ClaudeCodePSymphony) | package.json declares ISC; no root LICENSE detected | 📚 Study | Tiny TypeScript fork of OpenAI Symphony that uses GitHub Issues labels as the work queue and Claude Code `--print` as the agent runner; useful pattern source, but high-trust `bypassPermissions`, raw shell hooks, prompt-injection exposure, no tests/CI, and dependency advisories block deployment |
| [needle.md](reviews/needle.md) | [cactus-compute/needle](https://github.com/cactus-compute/needle) | MIT | ⚠️ Interesting | Tiny 26M-parameter JAX/Flax function-calling model and local finetune toolkit for on-device agents, with constrained decoding and open weights, but pickle checkpoints, side-effectful setup scripts, sparse tests, and unreproduced benchmark claims keep it experimental |
| [harper.md](reviews/harper.md) | [Automattic/harper](https://github.com/Automattic/harper) | Apache-2.0 | ✅ Deploy candidate | Offline Rust/WASM grammar and spell checker with local-first core, LSP/CLI/JS surfaces, browser/editor integrations, code-aware prose extraction, and strong Rust test coverage |
| [adhd.md](reviews/adhd.md) | [UditAkhourii/adhd](https://github.com/UditAkhourii/adhd) | MIT | ✅ Deploy candidate | Coding-agent skill, CLI, and TypeScript library for isolated divergent ideation branches with separate critic scoring, trap pruning, clustering, and deepening |
| [langchain-skills.md](reviews/langchain-skills.md) | [langchain-ai/langchain-skills](https://github.com/langchain-ai/langchain-skills) | MIT declared in plugin metadata; no root LICENSE detected | ✅ Deploy candidate | Official LangChain Agent Skills pack for LangChain, LangGraph, Deep Agents, Managed Deep Agents, RAG, HITL, persistence, eval engineering, and a tested TypeScript swarm helper |
| [buzz.md](reviews/buzz.md) | [block/buzz](https://github.com/block/buzz) | Apache-2.0 | ✅ Deploy candidate | Self-hostable human-and-agent workspace built on a signed Nostr relay, with Rust services, Tauri/React/Flutter clients, JSON-first CLI, workflows, git events, media, search, audit, and serious tenant-boundary design |
| [kimi-code.md](reviews/kimi-code.md) | [MoonshotAI/kimi-code](https://github.com/MoonshotAI/kimi-code) | MIT | ✅ Deploy candidate | Moonshot AI's open-source TypeScript coding-agent platform with TUI, web/server mode, ACP, MCP, plugins, subagents, goal mode, permissions, native packaging, and active releases |
| [clodex-ide.md](reviews/clodex-ide.md) | [mereyabdenbekuly-ctrl/clodex-ide](https://github.com/mereyabdenbekuly-ctrl/clodex-ide) | AGPL-3.0-only | ⚠️ Interesting | Local-first Electron/TypeScript agentic IDE for durable software work, with process isolation, Guardian/approval packages, MCP/model/provider surfaces, release evidence automation, and strong zero-trust design, but very fresh and currently unsigned technical-preview software |
| [mindwalk.md](reviews/mindwalk.md) | [cosmtrek/mindwalk](https://github.com/cosmtrek/mindwalk) | MIT | ✅ Deploy candidate | Local Go/React/Three visualizer for replaying Claude Code and Codex sessions over deterministic repo maps, with trace/citymap/report artifacts, subagent lenses, sealed optional judge, schemas, CI, releases, and passing local tests |
| [exercises-dataset.md](reviews/exercises-dataset.md) | [hasaneyldrm/exercises-dataset](https://github.com/hasaneyldrm/exercises-dataset) | MIT for code/data/text; media excluded | ⚠️ Interesting | Static 1,324-exercise fitness dataset with JSON Schema, thumbnails/GIFs, 10-language instructions, and browser/setup tools; useful app seed with Gym visual media-license caveats and a French-omitting SQL exporter |
| [wallbreaker.md](reviews/wallbreaker.md) | [JailbrokenAI/wallbreaker](https://github.com/JailbrokenAI/wallbreaker) | AGPL-3.0-or-later | ⚠️ Interesting | Python LLM red-team harness with attacker/target/judge roles, Textual/Rich CLI, provider adapters, MCP bridge, optional dashboard, automated campaigns, reporting/regrading/export, and strong authorized-lab-only caveats |
| [gitnexus.md](reviews/gitnexus.md) | [abhigyanpatwari/GitNexus](https://github.com/abhigyanpatwari/GitNexus) | PolyForm Noncommercial 1.0.0 | ✅ Deploy candidate | Local-first code intelligence engine for AI coding agents with a LadybugDB knowledge graph, MCP/CLI tools, React graph explorer, broad Tree-sitter language coverage, read-only mode, and serious CI/security posture |
| [airgorah.md](reviews/airgorah.md) | [martin-olivier/airgorah](https://github.com/martin-olivier/airgorah) | MIT | ⚠️ Interesting | Rust/GTK4 Linux GUI for authorized WiFi auditing over aircrack-ng tools; useful operator wrapper with CI/package signals, but root privileges, service stopping, shell composition, and sparse tests make it a dedicated-lab tool rather than a casual install |
| [corey-ganim-ai-tools-assessment.md](reviews/corey-ganim-ai-tools-assessment.md) | [Corey Ganim AI Tools Assessment thread](https://x.com/coreyganim/status/2077788148285792662) | N/A | ⚠️ Interesting | X thread outlining a paid AI tools assessment: useful diagnosis-before-cure intake pattern, but anecdotal revenue claims and no inspectable public template behind the linked landing page |
| [superpowers-v.md](reviews/superpowers-v.md) | [procoders/superpowers-v](https://github.com/procoders/superpowers-v) | MIT | ✅ Deploy candidate | Claude Code plugin layering deterministic multi-model orchestration on Superpowers, with pre-flights, manifest partitioning, multi-backend dispatch, git-derived scope gates, crash-resumable epics, local memory, and broad selftests |
| [tool-eval-bench.md](reviews/tool-eval-bench.md) | [SeraphimSerapis/tool-eval-bench](https://github.com/SeraphimSerapis/tool-eval-bench) | MIT | ✅ Deploy candidate | Deterministic Python benchmark for tool-calling quality across OpenAI-compatible serving stacks, with 69+ scenarios, hard mode, safety-capped scoring, trace reports, plugin benchmarks, Docker/CI, and 2107 passing local tests |
| [rodmanai-50-github-repos.md](reviews/rodmanai-50-github-repos.md) | [50 GitHub Repos Every Developer Needs](https://x.com/RodmanAi/status/2077772397328138401) | N/A | ⚠️ Interesting | X post listing 50 developer and AI GitHub repositories; useful as discovery backlog fuel, but too shallow and unsorted to treat as a vetted recommendation registry |
| [ai-marketing-skills.md](reviews/ai-marketing-skills.md) | [ericosiu/ai-marketing-skills](https://github.com/ericosiu/ai-marketing-skills) | MIT | ⚠️ Interesting | Large Claude Code-style marketing/revenue skill pack with expert panels, growth experiment tracking, content scoring, closed-loop analytics, sales/outbound scripts, local telemetry, and a noisy/inactive safety posture that needs pruning before use |
| [codex-orchestration.md](reviews/codex-orchestration.md) | [Cjbuilds/Codex-Orchestration](https://github.com/Cjbuilds/Codex-Orchestration) | MIT | ✅ Deploy candidate | Codex plugin for persistent Planner/Advisor/Executor model routing, reversible setup/status/disable, bounded Claude Fable 5 bridge, custom-agent routes, and strong contract/config tests |
| [whathappened.md](reviews/whathappened.md) | [kunchenguid/whathappened](https://github.com/kunchenguid/whathappened) | MIT | ⚠️ Interesting | Grok Build-only Agent Skill for X/Twitter event briefings with hard X-tool gates, adaptive recency windows, Top/Latest/Semantic/first-party/debate search lanes, opinion maps, receipts, and failure-mode guidance |
| [bolt-slides.md](reviews/bolt-slides.md) | [stackblitz/bolt-slides](https://github.com/stackblitz/bolt-slides) | MIT | ✅ Deploy candidate | React/Vite presentation deck engine plus agent-facing authoring skill for responsive, interactive web-app slides, with click-builds, presenter mode, annotations, component kit, and a useful locked-engine/editable-content boundary |
| [personal-claude-code-cheatsheet.md](reviews/personal-claude-code-cheatsheet.md) | [sants2001/personal-claude-code-cheatsheet](https://github.com/sants2001/personal-claude-code-cheatsheet) | **no license**; README claims MIT | ⚠️ Interesting | Tiny Claude Code skill that generates a personalized offline HTML cheat sheet from the live session skill list and local CLAUDE.md files; useful live-environment-reflection pattern, but no license file, tests, manifest, or deterministic generator |
| [compound-engineering-plugin.md](reviews/compound-engineering-plugin.md) | [EveryInc/compound-engineering-plugin](https://github.com/EveryInc/compound-engineering-plugin) | MIT | ✅ Deploy candidate | Cross-harness agent workflow plugin for ideation, requirements, planning, implementation, review, PR feedback, and compounding learnings, with Bun/TypeScript converters, native plugin manifests, strong contract tests, and careful installer safety |
| [colibri.md](reviews/colibri.md) | [JustVugg/colibri](https://github.com/JustVugg/colibri) | Apache-2.0 | ✅ Deploy candidate | Local MoE inference platform for GLM-5.2, Inkling, Kimi K3, and OLMoE, with NVMe/RAM/VRAM expert placement, CUDA/HIP/Metal/Vulkan paths, OpenAI/Anthropic-compatible APIs, release archive verification, dashboard, and web lockfile/audit caveats |
| [omnigent.md](reviews/omnigent.md) | [omnigent-ai/omnigent](https://github.com/omnigent-ai/omnigent) | Apache-2.0 | ✅ Deploy candidate | Python meta-harness/control plane for Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, Kimi, Qwen, Copilot, Antigravity, and custom YAML agents, with FastAPI server, sharing, policies, sandboxing, credential proxying, cloud sandbox providers, and broad CI |
| [superpowers.md](reviews/superpowers.md) | [obra/superpowers](https://github.com/obra/superpowers) | MIT | ✅ Deploy candidate | Multi-harness Agent Skills methodology for coding agents with brainstorming, planning, TDD, file-backed subagent-driven development, skeptical task review, Codex marketplace packaging, and strong AI-contribution guardrails |
| [destructive-command-guard.md](reviews/destructive-command-guard.md) | [Dicklesworthstone/destructive_command_guard](https://github.com/Dicklesworthstone/destructive_command_guard) | MIT-style with OpenAI/Anthropic rider | ⚠️ Interesting | Rust command guard for AI coding agents that blocks destructive shell/git/cloud/db/Kubernetes commands before execution, with agent-specific hook protocols, modular rule packs, heredoc/inline-script scanning, scan/SARIF modes, broad tests, and a restrictive/non-OSI license rider |
| [kill-ai-slop.md](reviews/kill-ai-slop.md) | [yetone/kill-ai-slop](https://github.com/yetone/kill-ai-slop) | **no license** | ⚠️ Interesting | Static field guide and Agent Skill for identifying AI-generated visual/copy tics, with a 23-item taxonomy, before/after demos, dependency-free read-only Node scanner, triage-before-edit workflow, and caveats around missing license/tests plus current Astro audit advisories |
| [lmcache.md](reviews/lmcache.md) | [LMCache/LMCache](https://github.com/LMCache/LMCache) | Apache-2.0 | ✅ Deploy candidate | KV-cache management layer for LLM inference with vLLM/SGLang/TensorRT-LLM integrations, multiprocess serving, tiered CPU/disk/remote/P2P storage, CacheBlend non-prefix reuse, observability, controller APIs, Docker/Kubernetes deployment, and privileged-infra caveats |
| [texts-to-transformer.md](reviews/texts-to-transformer.md) | [Doriandarko/texts-to-transformer](https://github.com/Doriandarko/texts-to-transformer) | MIT | ✅ Deploy candidate | Local Apple Silicon pipeline for training a tiny from-scratch iMessage-style Transformer with read-only SQLite snapshotting, HMAC pseudonyms, chronological guard-band splits, train-only tokenizer, MLX training, aggregate-only evaluation, and no-send terminal reply generation |
| [emilkowalski-skills.md](reviews/emilkowalski-skills.md) | [emilkowalski/skills](https://github.com/emilkowalski/skills) | MIT | ✅ Deploy candidate | Compact design-engineering skill pack for AI agents with strict animation-review standards, concrete easing/duration/origin/accessibility rules, motion vocabulary, and Apple-style fluid-interface guidance |
| [poormansadvisor.md](reviews/poormansadvisor.md) | [leighstillard/poormansadvisor](https://github.com/leighstillard/poormansadvisor) | README says MIT; no LICENSE file | ⚠️ Interesting | Single-file Claude Code advisor-consult skill with useful stuck-loop and architecture-decision triggers, but weak packaging, no validation, brittle model/subagent assumptions, and an unsafe verbatim-return rule |
| [fable-advisor.md](reviews/fable-advisor.md) | [DannyMac180/fable-advisor](https://github.com/DannyMac180/fable-advisor) | MIT | ⚠️ Interesting | Claude Code plugin and skill for architect-as-orchestrator model routing: top model keeps judgment/specs/verification while Sonnet, Opus, and Codex lanes handle implementation volume; useful contracts, but model/account assumptions and Codex workspace authority need care |
| [dctanner-cook-skill.md](reviews/dctanner-cook-skill.md) | [dctanner cook Skill](https://gist.github.com/dctanner/54c57da4a94a24e71df6281f487f51e1) | **no license** | ⚠️ Interesting | Tiny agent coding workflow skill with a useful plan-approve-implement-review-smoke-test shape, but unsafe defaults around `codex --yolo`, environment-specific model names, real requests, and commit/push authority |
| [tencentdb-agent-memory.md](reviews/tencentdb-agent-memory.md) | [TencentCloud/TencentDB-Agent-Memory](https://github.com/TencentCloud/TencentDB-Agent-Memory) | MIT text; GitHub NOASSERTION | ✅ Deploy candidate | Local-first OpenClaw/Hermes memory plugin with L0/L1/L2/L3 progressive memory, SQLite/sqlite-vec storage, optional Tencent Cloud VectorDB, hybrid BM25/vector/RRF retrieval, symbolic Mermaid context offload, Gateway API, and important caveats around CI, postinstall patching, and gateway auth |
| [awesome-deepseek-agent.md](reviews/awesome-deepseek-agent.md) | [deepseek-ai/awesome-deepseek-agent](https://github.com/deepseek-ai/awesome-deepseek-agent) | **no license** | 📚 Study | DeepSeek integration guide catalog covering agent/coding tools with bilingual setup docs, current V4 model-name rules, 1M context guidance, reasoning-effort notes, `reasoning_content`/`tool_choice` compatibility gotchas, and manual-review caveats |
| [open-multi-agent.md](reviews/open-multi-agent.md) | [open-multi-agent/open-multi-agent](https://github.com/open-multi-agent/open-multi-agent) | MIT | ✅ Deploy candidate | TypeScript multi-agent orchestration framework with goal-to-DAG planning, deterministic scheduling, plan preview/replay, checkpoint/resume, default-deny tools, model routing, broad provider support, and a useful scaffolder |
| [agent-skills.md](reviews/agent-skills.md) | [addyosmani/agent-skills](https://github.com/addyosmani/agent-skills) | MIT | ✅ Deploy candidate | Production-engineering skill pack for AI coding agents with 24 lifecycle skills, 8 slash commands, specialist review/security/test/perf personas, anti-rationalization gates, command parity validation, plugin install CI, and optional hooks |
| [agent-kernel-arena.md](reviews/agent-kernel-arena.md) | [AMD-AGI/AgentKernelArena](https://github.com/AMD-AGI/AgentKernelArena) | Apache-2.0 | ✅ Deploy candidate | AMD ROCm GPU-kernel optimization benchmark arena for comparing Cursor, Claude Code, Codex, and custom agents on isolated HIP/Triton/Torch2HIP/FlyDSL tasks with centralized compile/correctness/performance scoring, multi-GPU workers, held-out shape evals, and privileged-lab caveats |
| [karakeep.md](reviews/karakeep.md) | [karakeep-app/karakeep](https://github.com/karakeep-app/karakeep) | AGPL-3.0 | ✅ Deploy candidate | Self-hosted bookmark-everything/read-it-later system with Next.js, SQLite, Meilisearch, workers, crawler/archive/OCR/video pipelines, AI tagging/summarization, browser/mobile clients, CLI, SDK, MCP, scoped API keys, and Docker deployment |
| [codebase-to-course.md](reviews/codebase-to-course.md) | [zarazhangrui/codebase-to-course](https://github.com/zarazhangrui/codebase-to-course) | **no license** | 📚 Study | Claude Code skill for generating static interactive codebase courses with product-first modules, exact code-to-English translations, visual density rules, scenario quizzes, glossary tooltips, module briefs, and prebuilt HTML/CSS/JS assets |
| [xint.md](reviews/xint.md) | [0xNyk/xint](https://github.com/0xNyk/xint) | MIT | ✅ Deploy candidate | Agent-native X/Twitter intelligence CLI and MCP server with Bun/TypeScript, X API and xAI integrations, policy-gated read/engagement/moderation modes, cost tracking, local exports, TUI, webhook/server hardening, and current CI/security guards |
| [council-of-high-intelligence.md](reviews/council-of-high-intelligence.md) | [0xNyk/council-of-high-intelligence](https://github.com/0xNyk/council-of-high-intelligence) | MIT | ✅ Deploy candidate | Multi-persona, multi-provider deliberation skill with 18 reasoning seats, blind-first analysis, anonymized cross-examination, method diversity, confidence-weighted stances, chairman synthesis, and Claude/Codex/Gemini/opencode packaging |
| [mission-control.md](reviews/mission-control.md) | [builderz-labs/mission-control](https://github.com/builderz-labs/mission-control) | MIT | ✅ Deploy candidate | Self-hosted AI agent operations dashboard/control plane with Next.js, SQLite, RBAC, task/review gates, runtime dispatch, skills, memory, costs, security scans, OpenAPI parity, and serious test/CI posture |
| [awesome-hermes-agent.md](reviews/awesome-hermes-agent.md) | [0xNyk/awesome-hermes-agent](https://github.com/0xNyk/awesome-hermes-agent) | CC BY 4.0 | 📚 Study | Large curated Hermes Agent ecosystem map with 189 maturity-tagged skills, plugins, tools, memory providers, integrations, deployment resources, operational playbooks, and blueprint bundles; useful for discovery, not as a verified registry |
| [codesherpa.md](reviews/codesherpa.md) | [farhankhwaja/codesherpa](https://github.com/farhankhwaja/codesherpa) | Apache-2.0 | ✅ Deploy candidate | Local git-native structural memory for coding agents with blob-keyed incremental indexing, tree-sitter chunks, SQLite/FTS5/sqlite-vec storage, hybrid retrieval, compact MCP tools, local analytics, and candid eval caveats |
| [dzhng-skills.md](reviews/dzhng-skills.md) | [dzhng/skills](https://github.com/dzhng/skills) | MIT | ✅ Deploy candidate | Portable Agent Skills catalog for software-factory workflows: unknown mapping, living spec slice graphs, iterative implementation passes, visual comparison gates, refactor cleanup, second-agent review, and skill authoring/eval discipline |
| [meetily.md](reviews/meetily.md) | [Zackriya-Solutions/meetily](https://github.com/Zackriya-Solutions/meetily) | MIT | ✅ Deploy candidate | Local-first Tauri meeting assistant with Rust audio capture, Whisper/Parakeet transcription, import/retranscription, SQLite meeting folders, local or cloud summary providers, and caveats around broad native permissions and model provenance |
| [traycer.md](reviews/traycer.md) | [traycerai/traycer](https://github.com/traycerai/traycer) | Apache-2.0 | ✅ Deploy candidate | Desktop multi-agent coding workspace with React/Electron GUI, Bun/Nx TypeScript monorepo, signed local host lifecycle, per-method versioned RPC, Yjs-backed Epic state, terminal agents, collaboration, and strong CI/security hygiene |
| [outline.md](reviews/outline.md) | [outline/outline](https://github.com/outline/outline) | BSL-1.1 | ✅ Deploy candidate | Mature collaborative team knowledge base with React/Vite/MobX frontend, Koa/Postgres/Redis backend, ProseMirror/Hocuspocus realtime editing, large integration surface, Docker self-hosting, strong CI, and a scope-aware MCP endpoint |
| [arcane.md](reviews/arcane.md) | [getarcaneapp/arcane](https://github.com/getarcaneapp/arcane) | BSD-3-Clause | ✅ Deploy candidate | Self-hosted Docker/Compose control plane with Go/Huma backend, SvelteKit UI, granular RBAC, OIDC/API keys, edge agents, GitOps syncs, vulnerability scans, signed multi-arch releases, and Docker-socket deployment caveats |
| [use-codex-skill.md](reviews/use-codex-skill.md) | [Use Codex Skill](https://public.my-agent-04eee268.sandbox.dev/skills/use-codex.md) | N/A | ⚠️ Interesting | Skill document for delegating context-heavy work to Codex CLI subagents; useful parent/subagent synthesis pattern, but risky defaults around shell-spawned agents, broad execution flags, generic temp files, and weak validation guardrails |
| [cve-env.md](reviews/cve-env.md) | [gadievron/cve-env](https://github.com/gadievron/cve-env) | MIT | ⚠️ Interesting | Agentic CVE-to-Docker lab builder that researches affected versions, pulls or builds pre-patch environments, launches them with hardened localhost-only defaults, and requires both version proof and benign functional smoke before marking success |
| [agentsonar.md](reviews/agentsonar.md) | [knostic/AgentSonar](https://github.com/knostic/AgentSonar) | **no license** | ⚠️ Interesting | Go endpoint network monitor for shadow-AI detection, joining DNS/TLS/SNI/streaming traffic to local process/domain pairs with heuristic AI scoring, SQLite events, overrides, external classifiers, Sigma export, and partial router/mirror support via PID0 mode |
| [exploitation-validator.md](reviews/exploitation-validator.md) | [gadievron/exploitation-validator](https://github.com/gadievron/exploitation-validator) | MIT | 🔧 Harvest | Prompt-only staged exploitability validation pipeline for LLM sub-agents, with inventory, one-shot proof attempts, systematic attack-path documents, source sanity checks, ruling filters, and final self-review |
| [os-june.md](reviews/os-june.md) | [open-software-network/os-june](https://github.com/open-software-network/os-june) | MIT | ✅ Deploy candidate | Private Tauri desktop AI assistant for meeting notes, dictation, and local Hermes agent work, with saved-audio-first capture, SQLite local state, June API model proxying, TEE attestation docs, and mature macOS/Windows release workflows |
| [fractal.md](reviews/fractal.md) | [Trampoline-AI/fractal](https://github.com/Trampoline-AI/fractal) | MIT | ✅ Deploy candidate | Recursive Language Model terminal coding agent with predict-rlm, Docker SBX direct workspace mounts, provider setup, session persistence, TUI, headless JSON delegation mode, installable agent skill, and 270 passing tests |
| [mineru.md](reviews/mineru.md) | [opendatalab/MinerU](https://github.com/opendatalab/MinerU) | MinerU Open Source License | ✅ Deploy candidate | Multi-format document parsing engine for PDFs, images, DOCX, PPTX, and XLSX with Markdown/JSON outputs, hybrid/VLM/pipeline backends, FastAPI tasks, router deployment, Gradio, Docker, and public-bind SSRF guard |
| [handy.md](reviews/handy.md) | [cjpais/Handy](https://github.com/cjpais/Handy) | MIT | ✅ Deploy candidate | Cross-platform local speech-to-text desktop app with Rust/Tauri, shortcut-driven recording, multi-engine local ASR model management, VAD, active-app paste, optional transcript post-processing, signed releases, and candid Linux/Wayland caveats |
| [improve.md](reviews/improve.md) | [shadcn/improve](https://github.com/shadcn/improve) | MIT | ✅ Deploy candidate | Agent Skill for read-only codebase audits that turns vetted findings into commit-stamped, self-contained implementation plans for separate executor agents or humans |
| [kulshan.md](reviews/kulshan.md) | [azz-kikkr/kulshan](https://github.com/azz-kikkr/kulshan) | Apache-2.0 | ✅ Deploy candidate | Local-first read-only AWS FinOps and posture audit CLI with ten opt-in packs, explicit IAM policies, local HTML/JSON/SARIF/CSV reports, PII redaction, Cost Explorer cost warnings, and agent-readable output |
| [dynoxide.md](reviews/dynoxide.md) | [nubo-db/dynoxide](https://github.com/nubo-db/dynoxide) | MIT OR Apache-2.0 | ✅ Deploy candidate | Fast embeddable DynamoDB emulator in Rust over SQLite with HTTP, Rust library, Docker, GitHub Action, MCP server, WASM preview, conformance-driven compatibility, snapshots, and clear local-agent security controls |
| [drawio-skill.md](reviews/drawio-skill.md) | [Agents365-ai/drawio-skill](https://github.com/Agents365-ai/drawio-skill) | MIT | ✅ Deploy candidate | Agent Skill for editable draw.io diagram generation with codebase-to-diagram importers, Graphviz auto-layout, exact vendor shape search, AI/LLM brand logos, deterministic XML validation, and PNG export repair |
| [visual-explainer.md](reviews/visual-explainer.md) | [nicobailon/visual-explainer](https://github.com/nicobailon/visual-explainer) | MIT | ⚠️ Interesting | Multi-harness Agent Skill for turning complex terminal output into self-contained HTML diagrams, diff reviews, plan reviews, project recaps, slide decks, and dense tables, now with a Pi render tool and strong Mermaid/overflow guidance |
| [htmlbin-cli.md](reviews/htmlbin-cli.md) | [utsengar/htmlbin-cli](https://github.com/utsengar/htmlbin-cli) | MIT | ⚠️ Interesting | TypeScript CLI for publishing single-file HTML artifacts to public htmlbin.dev drops, GitHub Pages, or Cloudflare Pages with agent JSON output, stable metadata upserts, pattern-guided generation, and CI preview workflows |
| [palpatine.md](reviews/palpatine.md) | [NovusEdge/palpatine](https://github.com/NovusEdge/palpatine) | Non-standard SPL; manifest says MIT | ❌ Pass | Claude Code skill/plugin bundle for Robert Greene-style power, seduction, war, defense, wargaming, and bounded subagent orchestration; behavior layer is intentionally manipulative and license metadata conflicts |
| [hunk.md](reviews/hunk.md) | [modem-dev/hunk](https://github.com/modem-dev/hunk) | MIT | ✅ Deploy candidate | Review-first terminal diff viewer for agent-authored changesets with OpenTUI/React rendering, Git/jj/Sapling loaders, inline agent annotations, loopback session control, reusable TUI components, and strong package/test coverage |
| [understand-anything.md](reviews/understand-anything.md) | [Egonex-AI/Understand-Anything](https://github.com/Egonex-AI/Understand-Anything) | MIT | ✅ Deploy candidate | Multi-platform codebase/knowledge-base graph plugin with deterministic scanning, semantic batching, import maps, LLM-enriched nodes, graph validation, local dashboard exploration, diff impact analysis, and dependency hygiene caveats |
| [ponytail.md](reviews/ponytail.md) | [DietrichGebert/ponytail](https://github.com/DietrichGebert/ponytail) | MIT | ✅ Deploy candidate | Portable lazy-senior-dev behavior layer for coding agents with canonical skill source, npm packaging, many host adapters, lifecycle hooks, MCP surface, generated OpenClaw skills, drift tests, Codex/OpenCode/Pi/Qoder compatibility fixes, and agentic benchmark evidence |
| [storm.md](reviews/storm.md) | [stanford-oval/storm](https://github.com/stanford-oval/storm) | MIT | 📚 Study | Stanford OVAL knowledge curation engine for perspective-guided retrieval, staged outline/report generation, Co-STORM collaborative discourse, and citation-preserving research artifacts |
| [knowledge-catalog.md](reviews/knowledge-catalog.md) | [GoogleCloudPlatform/knowledge-catalog](https://github.com/GoogleCloudPlatform/knowledge-catalog) | Apache-2.0 | ⚠️ Interesting | Google Cloud sample repo for Open Knowledge Format, metadata-as-code catalog snapshots, agentic data-documentation enrichment, MCP catalog tooling, and static graph viewing |
| [context-firewall.md](reviews/context-firewall.md) | [nik1t7n/context-firewall](https://github.com/nik1t7n/context-firewall) | Apache-2.0 | ✅ Deploy candidate | Rust local-first command-output firewall for coding agents with deterministic reducers, local raw artifacts, SQLite spans, MCP run/show/receipt tools, policy gates, dedupe, and guarded exact-output retrieval |
| [oz-for-oss.md](reviews/oz-for-oss.md) | [warpdotdev/oz-for-oss](https://github.com/warpdotdev/oz-for-oss) | MIT | ⚠️ Interesting | Webhook-driven GitHub App control plane for Oz agents with explicit event routing, Vercel KV run state, cron-side result application, repo-local skill companions, and strong tests |
| [esperanto.md](reviews/esperanto.md) | [lfnovo/esperanto](https://github.com/lfnovo/esperanto) | MIT | ✅ Deploy candidate | Lightweight Python direct-HTTP provider-normalization library for LLMs, embeddings, reranking, STT, and TTS, with OpenAI-compatible profiles, common typed responses, release-gated real API tests, and a strong mocked parity suite |
| [fabro.md](reviews/fabro.md) | [fabro-sh/fabro](https://github.com/fabro-sh/fabro) | MIT | ✅ Deploy candidate | Rust/React dark software factory for deterministic Graphviz-coded AI workflows with typed agent/command/human/parallel nodes, model routing, sandbox providers, checkpoints, OpenAPI/SSE control plane, and explicit single-tenant security caveats |
| [agent-scripts.md](reviews/agent-scripts.md) | [steipete/agent-scripts](https://github.com/steipete/agent-scripts) | MIT | 📚 Study | Canonical shared coding-agent operations repo with routeable skills, pointer-style shared instructions, docs metadata, small portable helpers, and validation for skill front matter |
| [agent-skills-tmchow.md](reviews/agent-skills-tmchow.md) | [tmchow/agent-skills](https://github.com/tmchow/agent-skills) | MIT | ✅ Deploy candidate | Cross-runtime Agent Skills catalog with Camofox, Chrome DevTools AXI, Clawpatch, and Illo skills; strong README/SKILL split, runtime metadata, scanner-safe credential rules, and checksum-gated binary asset repair |
| [rockyvoice.md](reviews/rockyvoice.md) | [Lagunaswift/RockyVoice](https://github.com/Lagunaswift/RockyVoice) | MIT code; fan-project character/audio caveats | 🔧 Harvest | Claude persona skill and local Hume TTS voice app with explicit off-switch, safety-clarity override, Stop-hook ingestion, Markdown-to-speech cleanup, SSE streaming, and Web Audio playback |
| [continual-learning-bench.md](reviews/continual-learning-bench.md) | [pgasawa/continual-learning-bench](https://github.com/pgasawa/continual-learning-bench) | Apache-2.0 | ✅ Deploy candidate | Python 3.13 benchmark harness for stateful AI agents with task/system contracts, reset-baseline gain metrics, trace artifacts, Dockerized Claude/Codex/Mem0/ACE/ICL systems, and substantial sequential tasks |
| [continual-learning-bench-1-0.md](reviews/continual-learning-bench-1-0.md) | [Continual Learning Bench 1.0](https://continual-learning-bench.com/news/cl-bench-1-0/) | N/A | ⚠️ Interesting | Benchmark release for stateful AI systems that should improve across related task sequences, with useful gain-over-stateless-baseline framing but a still-pending technical report |
| [no-mistakes.md](reviews/no-mistakes.md) | [kunchenguid/no-mistakes](https://github.com/kunchenguid/no-mistakes) | MIT | ✅ Deploy candidate | Go local git proxy for AI-assisted delivery with managed `git push no-mistakes` remote, daemon, SQLite run ledger, JSON-RPC/TUI/AXI control surfaces, disposable worktrees, agent review/test/fix/PR/CI pipeline, and sandbox/telemetry caveats |
| [lalamo.md](reviews/lalamo.md) | [trymirai/lalamo](https://github.com/trymirai/lalamo) | MIT | ⚠️ Interesting | Python/JAX model conversion and optimization toolchain for Uzu with typed model specs, Hugging Face import, quantization/compression support, safetensors runtime artifact export, broad model-family coverage, and heavyweight CI/coherence validation |
| [uzu.md](reviews/uzu.md) | [trymirai/uzu](https://github.com/trymirai/uzu) | MIT | ✅ Deploy candidate | Rust on-device AI inference engine with native Metal/CPU backends, model registry and download management, chat/classification/TTS sessions, generated Rust/Python/TypeScript/Swift/WASM bindings, CLI, OpenAI-compatible server, and strong privacy caveats around registry/telemetry defaults |
| [mq.md](reviews/mq.md) | [harehare/mq](https://github.com/harehare/mq) | MIT | ✅ Deploy candidate | Rust jq-like Markdown query language and CLI with structured selectors, update mode, multi-format IO, formatter, type checker, LSP, REPL, DAP, WASM/web API, editor integrations, Agent Skill docs, and passing workspace fmt/test/clippy validation |
| [qship.md](reviews/qship.md) | [3awny/qship](https://github.com/3awny/qship) | MIT | ✅ Deploy candidate | Claude Code/Codex CLI ticket-to-PR pipeline with 21 rendered skills, hook-enforced phase gates, disk-backed state, Codex review/implementation routing, installer/config schema, smoke tests, eval scaffold, and strong sandboxing caveats |
| [agent-coding.md](reviews/agent-coding.md) | [dsweet99/agent_coding](https://github.com/dsweet99/agent_coding) | **no license** | 📚 Study | Cursor-oriented coding-agent prompt workflow plus Python `malvin` CLI for implementing, reviewing, falsifying concerns, fixing, and repeating until exact `LGTM`, with run artifacts and a passing 75-test suite when explicit test deps are installed |
| [last30days-skill.md](reviews/last30days-skill.md) | [mvanhorn/last30days-skill](https://github.com/mvanhorn/last30days-skill) | MIT | ✅ Deploy candidate | Agent Skill and Python research engine for last-30-days social/market/developer/web synthesis across Reddit, X, YouTube, TikTok, HN, Polymarket, GitHub, Digg, Perplexity, and more, with planning, fan-out, scoring, clustering, HTML briefs, and a large passing test suite |
| [omniretrieval.md](reviews/omniretrieval.md) | [JinheonBaek/OmniRetrieval](https://github.com/JinheonBaek/OmniRetrieval) | MIT | 📚 Study | Research code for unified retrieval across heterogeneous sources, routing natural-language questions to text search, SQL, SPARQL, or Cypher backends while preserving source-native query/execution semantics |
| [tech-snacks.md](reviews/tech-snacks.md) | [ragnar-pwninskjold/tech-snacks](https://github.com/ragnar-pwninskjold/tech-snacks) | MIT | ⚠️ Interesting | Claude Code plugin skill library with Markdown skills, references, templates, research agents, and workflow-backed procedures for UI cloning, PRD-to-UX, CLAUDE.md mining, and React refactor campaigns |
| [mamma.md](reviews/mamma.md) | [cuevhv/mamma](https://github.com/cuevhv/mamma) | Custom non-commercial scientific research | ⚠️ Interesting | CVPR 2026 markerless multi-person motion-acquisition pipeline with capture loading, SAM/YOLO segmentation, MammaNet 2D landmarks, SMPL-X optimization, Rerun/overlay visualization, local Flask/React GUI, gated assets, and heavy CUDA/runtime caveats |
| [art.md](reviews/art.md) | [OpenPipe/ART](https://github.com/OpenPipe/ART) | Apache-2.0 | ✅ Deploy candidate | Agent Reinforcement Trainer framework for training multi-step LLM agents with GRPO over grouped trajectories, OpenAI-compatible clients, vLLM/LoRA backends, RULER LLM-judge rewards, MCP-RL examples, and local or managed GPU training paths |
| [self-revising-discovery-systems.md](reviews/self-revising-discovery-systems.md) | [Self-Revising Discovery Systems for Science](https://arxiv.org/abs/2606.01444) | CC BY-NC-ND 4.0 | 📚 Good reference | Category-theoretic framework for auditable agentic science systems, distinguishing retrieval, search, and discovery as typed regime transitions with provenance, gates, retained rejections, and residual content |
| [defending-code-reference-harness.md](reviews/defending-code-reference-harness.md) | [anthropics/defending-code-reference-harness](https://github.com/anthropics/defending-code-reference-harness) | Apache-2.0 | 🔧 Harvest | Claude-driven vulnerability discovery and remediation reference with Claude Code skills, Docker/gVisor sandboxing, ASAN target containers, PoC-producing find agents, fresh-container verification, semantic dedupe, structured reports, and executable patch grading |
| [walkthrough.md](reviews/walkthrough.md) | [austeane/walkthrough](https://github.com/austeane/walkthrough) | **no license** | 📚 Study | Python pipeline for turning Codex, Claude Code, and OpenCode session histories into evidence-backed offline HTML walkthroughs with provider-normalized events, projection, chunking, source refs, media handling, and quality gates |
| [cuda-agent.md](reviews/cuda-agent.md) | [CUDA Agent](https://arxiv.org/html/2602.24286v1) | N/A | ⚠️ Interesting | Research paper on large-scale agentic RL for CUDA kernel generation, combining synthesized operator tasks, hardened compile/profile feedback, a CUDA skill loop, robust rewards, and staged actor/critic warm-up |
| [oh-my-pi.md](reviews/oh-my-pi.md) | [can1357/oh-my-pi](https://github.com/can1357/oh-my-pi) | MIT | ✅ Deploy candidate | Terminal coding-agent harness with Bun/TypeScript and Rust native tools, hash-anchored edits, LSP/DAP, `xd://` mounted tools, subagents, local memory, encrypted collaboration, auth broker/gateway, provider adapters, and `robomp` GitHub automation |
| [pi-codex-goal.md](reviews/pi-codex-goal.md) | [fitchmultz/pi-codex-goal](https://github.com/fitchmultz/pi-codex-goal) | MIT | ✅ Deploy candidate | Pi extension for Codex-style goal tracking with session-native persistence, model-callable goal tools, hidden continuations, token/elapsed accounting, stale-work cleanup, recovery handling, and cross-platform packed-install smoke tests |
| [learn-quiz-skill-gist.md](reviews/learn-quiz-skill-gist.md) | [ThariqS/Learn Quiz gist](https://gist.github.com/ThariqS/1389dcdff9eba4789887a2211370f06b) | **no license** | 📚 Study | Compact tutoring skill prompt for teach-back learning loops, running comprehension checklists, incremental explanation, and quiz/debugger-based verification; useful as an opt-in coaching pattern, not a default agent behavior |
| [deepagents-with-langsmith.md](reviews/deepagents-with-langsmith.md) | [langchain-samples/deepagents-with-langsmith](https://github.com/langchain-samples/deepagents-with-langsmith) | **no license** | 📚 Study | Workshop repo for Deep Agents and LangSmith showing path-routed memory, subagents, AGENTS.md, skills, HITL interrupts, deployment, tracing, datasets, and trajectory evals; useful as reference material, not a fork target |
| [optimizerduck.md](reviews/optimizerduck.md) | [itsfatduck/optimizerDuck](https://github.com/itsfatduck/optimizerDuck) | GPL-3.0 | ⚠️ Interesting | Windows 10/11 WPF optimizer with reversible registry, service, scheduled-task, shell, power, startup, AppX, and cleanup operations, plus risk labels, localization, tests, CI, and system-mutation caveats |
| [pi-observational-memory.md](reviews/pi-observational-memory.md) | [elpapi42/pi-observational-memory](https://github.com/elpapi42/pi-observational-memory) | MIT | ✅ Deploy candidate | Pi extension for long-session continuity with background observation/reflection workers, append-only memory ledger entries, deterministic compaction summaries, recallable source evidence, and strong TypeScript/Vitest validation |
| [turbovec.md](reviews/turbovec.md) | [RyanCodrai/turbovec](https://github.com/RyanCodrai/turbovec) | MIT | ✅ Deploy candidate | Rust/Python local vector index built on TurboQuant with 2-4 bit embedding compression, SIMD search, stable IDs, kernel-level allowlist filtering, framework integrations, and strong correctness tests |
| [victor-mustar-glm-51-pi.md](reviews/victor-mustar-glm-51-pi.md) | [Victor Mustar on GLM-5.1 + Pi](https://x.com/victormustar/status/2063017894221591008) | N/A | ⚠️ Interesting | Practitioner signal that GLM-5.1 plus Pi may be a strong open coding-agent stack, with an anecdotal Three.js game eval emphasizing self-instrumentation, telemetry, and math-backed debugging |
| [agentic-inbox.md](reviews/agentic-inbox.md) | [cloudflare/agentic-inbox](https://github.com/cloudflare/agentic-inbox) | Apache-2.0 | ⚠️ Interesting | Cloudflare Workers email client and AI assistant with Email Routing inbound, Email Service outbound, Durable Object SQLite mailboxes, R2 attachments, Cloudflare Access auth, Workers AI draft agent, MCP tools, and mailbox-authorization caveats |
| [financetoolkit.md](reviews/financetoolkit.md) | [JerBouma/FinanceToolkit](https://github.com/JerBouma/FinanceToolkit) | MIT | ✅ Deploy candidate | Python financial analysis toolkit with transparent formulas for statements, ratios, valuation models, options, performance, risk, technicals, fixed income, economics, discovery, and portfolios, with external-data and snapshot-test caveats |
| [agentic-stack.md](reviews/agentic-stack.md) | [codejunkie99/agentic-stack](https://github.com/codejunkie99/agentic-stack) | Apache-2.0 | ⚠️ Interesting | Portable `.agent` memory/skills/protocols layer with manifest-driven adapters, local data-layer and flywheel exports, transfer bundles, safe install/remove/upgrade tooling, and a current Mission Control syntax regression in tests |
| [surya.md](reviews/surya.md) | [datalab-to/surya](https://github.com/datalab-to/surya) | Apache-2.0 code; modified OpenRAIL-M model license | ✅ Deploy candidate | Document-intelligence toolkit for OCR, layout analysis, reading order, and table recognition with a shared 650M VLM, vLLM/llama.cpp backends, GGUF support, structured JSON/HTML outputs, and strong local-runtime caveats |
| [minicpm5-1b.md](reviews/minicpm5-1b.md) | [openbmb/MiniCPM5-1B](https://huggingface.co/openbmb/MiniCPM5-1B) | Apache-2.0 | ⚠️ Interesting | Compact 1.08B local/on-device LLM with standard Llama architecture, 131k context, hybrid Think/No-Think chat template, XML-style tool calling, BF16/GGUF/MLX release variants, and deployment cookbooks |
| [skillspector.md](reviews/skillspector.md) | [NVIDIA/SkillSpector](https://github.com/NVIDIA/SkillSpector) | Apache-2.0 | ✅ Deploy candidate | Security scanner for AI agent skills with LangGraph analyzer fan-out, static/AST/YARA/OSV checks, MCP least-privilege and tool-poisoning rules, optional LLM filtering, SARIF output, Docker packaging, and strong Python 3.12 test coverage |
| [supermemory.md](reviews/supermemory.md) | [supermemoryai/supermemory](https://github.com/supermemoryai/supermemory) | MIT | ✅ Deploy candidate | Memory/context platform for AI agents with hosted app/API, MCP server, SDK/tool wrappers, browser/Raycast extensions, memory graph visualization, docs, benchmarks, and enterprise self-hosting path |
| [zero-trust-for-ai-agents.md](reviews/zero-trust-for-ai-agents.md) | [Zero Trust for AI Agents](https://cdn.prod.website-files.com/6889473510b50328dbb70ae6/6a1611a04085d7cd3dadc924_Claude-eBook-Zero-Trust-for-AI-Agents-05182026.pdf) | N/A | ✅ Act on this | Anthropic/Claude enterprise guide translating Zero Trust into agent controls: cryptographic identity, least agency, tool allowlists, short-lived credentials, sandboxing, memory integrity, observability, and human-gated incident response |
| [ad-filter-list.md](reviews/ad-filter-list.md) | [luisfelipess/ad-filter-list](https://github.com/luisfelipess/ad-filter-list) | MIT | ⚠️ Interesting | Small Python blocklist compiler that fetches upstream DNS/ad-block sources, applies allow/block overrides, validates/dedupes domains, publishes tiered hosts/adblock/RPZ/dnsmasq/Unbound outputs, and commits provenance reports |
| [stirling-pdf.md](reviews/stirling-pdf.md) | [Stirling-Tools/Stirling-PDF](https://github.com/Stirling-Tools/Stirling-PDF) | Mixed MIT / Stirling PDF User License | ✅ Deploy candidate | Self-hosted PDF platform with browser, desktop, API, workflow automation, OCR, redaction, signing, conversion, dependency-gated endpoints, Docker packaging, and mixed-license reuse caveats |
| [tracebase.md](reviews/tracebase.md) | [ssreeni1/tracebase](https://github.com/ssreeni1/tracebase) | MIT | ✅ Deploy candidate | Local-first trace capture and inspection for Codex/Claude sessions with encrypted raw blobs, redacted SQLite/FTS search, localhost dashboard/API, run intelligence, redacted exports, read-only MCP, and passing full test suite |
| [muse-autoskill.md](reviews/muse-autoskill.md) | [MUSE-Autoskill](https://arxiv.org/abs/2605.27366) | N/A | 📚 Good reference | Research paper proposing lifecycle-managed, test-gated, memory-bearing agent skills with SkillsBench evidence for skill reuse and cross-agent transfer |
| [activegraph-gbrain-bridge.md](reviews/activegraph-gbrain-bridge.md) | [apprentice-labs/activegraph-gbrain-bridge](https://github.com/apprentice-labs/activegraph-gbrain-bridge) | MIT | ⚠️ Interesting | Proof-of-concept bridge that keeps durable knowledge/schema in GBrain and runtime provenance/replay/fork/writeback policy in ActiveGraph, with fixture-first demos, redaction, compatibility ledgers, and dry-run mutation discipline |
| [flashlib.md](reviews/flashlib.md) | [FlashML-org/flashlib](https://github.com/FlashML-org/flashlib) | Apache-2.0 | ⚠️ Interesting | Alpha GPU library for classical ML operators with Triton/CuteDSL kernels, sklearn-style APIs, tolerance-driven precision routing, benchmark/cost models, and an agent-friendly planning surface with import-boundary caveats |
| [kindle-ai-export.md](reviews/kindle-ai-export.md) | [transitive-bullshit/kindle-ai-export](https://github.com/transitive-bullshit/kindle-ai-export) | MIT code; exported content remains copyright/terms-bound | ⚠️ Interesting | Local Kindle Web Reader export pipeline using Patchright/Playwright rendered-page capture, OpenAI vision OCR, metadata-aware stitching, PDF/Markdown/EPUB export, and optional OpenAI/Unreal Speech audiobook generation |
| [webwright.md](reviews/webwright.md) | [microsoft/Webwright](https://github.com/microsoft/Webwright) | MIT | ✅ Deploy candidate | Minimal Python/Playwright browser-agent harness that makes long-horizon web tasks produce rerunnable scripts, screenshots, logs, self-reflection outputs, and host-agent skill/plugin contracts |
| [iii.md](reviews/iii.md) | [iii-hq/iii](https://github.com/iii-hq/iii) | Elastic-2.0 engine; Apache-2.0 SDKs/console/docs/skills | ✅ Deploy candidate | Live worker/function/trigger backend runtime with Rust engine, Node/Python/Rust SDKs, React console, RBAC-filtered discovery, queues, cron, state, streams, OpenTelemetry, Docker/release pipelines, and agent-readable skills |
| [gepa-viz.md](reviews/gepa-viz.md) | [modaic-ai/gepa-viz](https://github.com/modaic-ai/gepa-viz) | MIT declared; no root LICENSE detected | ⚠️ Interesting | Live GEPA prompt-optimization visualizer with Python callback streaming, SSE/static run modes, React/D3 candidate tree, prompt diffs, minibatch feedback, Pareto grids, and durable run.json traces |
| [dreamserver.md](reviews/dreamserver.md) | [Light-Heart-Labs/DreamServer](https://github.com/Light-Heart-Labs/DreamServer) | Apache-2.0 | ✅ Deploy candidate | Hardware-aware local AI appliance installer/control plane for LLM inference, Open WebUI, voice, agents, workflows, RAG/search, ComfyUI, privacy tooling, observability, and generated-secret localhost-first deployment |
| [xurl.md](reviews/xurl.md) | [xdevplatform/xurl](https://github.com/xdevplatform/xurl) | MIT | ✅ Deploy candidate | Official curl-like CLI for the X API with OAuth2/OAuth1/app auth, multi-app token storage, shortcut commands, media upload, streaming, webhook helpers, and an agent-safe Claude skill reference |
| [open-notebook.md](reviews/open-notebook.md) | [lfnovo/open-notebook](https://github.com/lfnovo/open-notebook) | MIT | ✅ Deploy candidate | Self-hosted NotebookLM alternative with FastAPI, Next.js, SurrealDB, multi-provider/local AI, explicit context controls, REST API, transformations, notes, search, podcast generation, and active June 2026 security/dependency maintenance |
| [claude-bughunter.md](reviews/claude-bughunter.md) | [elementalsouls/Claude-BugHunter](https://github.com/elementalsouls/Claude-BugHunter) | MIT | ⚠️ Interesting | Claude Code skill bundle for authorized bug bounty and external red-team workflows with scope gates, validation discipline, evidence hygiene, reporting templates, slash commands, and a deterministic Python CLI |
| [devils-advocate.md](reviews/devils-advocate.md) | [brandonsimpson/devils-advocate](https://github.com/brandonsimpson/devils-advocate) | MIT | ✅ Deploy candidate | Claude Code plugin for adversarial binary code/plan critique with file:line evidence, fix suggestions, context gates, standards discovery, session logs, and non-blocking commit/plan hooks |
| [comfyui-ltxvideo.md](reviews/comfyui-ltxvideo.md) | [Lightricks/ComfyUI-LTXVideo](https://github.com/Lightricks/ComfyUI-LTXVideo) | LTX-2 Community License | ⚠️ Interesting | Official ComfyUI custom-node and workflow pack for LTX-2/LTX-2.3 video generation, with IC-LoRA, HDR, lipdub, motion tracking, low-VRAM loaders, and heavyweight license/runtime caveats |
| [hermes-vault.md](reviews/hermes-vault.md) | [asimons81/hermes-vault](https://github.com/asimons81/hermes-vault) | MIT | ✅ Deploy candidate | Local-first encrypted credential broker for AI agents with policy-gated ephemeral env access, MCP tools/resources, OAuth lifecycle support, verifier plugins, and a localhost operator dashboard |
| [12-factor-agents.md](reviews/12-factor-agents.md) | [humanlayer/12-factor-agents](https://github.com/humanlayer/12-factor-agents) | Apache-2.0 code / CC BY-SA 4.0 content | 📚 Study | Production-agent principles guide: owned prompts/context, structured tool outputs, unified state, pause/resume APIs, human tool calls, explicit control flow, compact errors, and reducer-style agents |
| [vimax.md](reviews/vimax.md) | [HKUDS/ViMax](https://github.com/HKUDS/ViMax) | MIT | 📚 Study | Agentic video-generation orchestration pipeline that stages story, character, storyboard, frame, provider-backed video, and composition artifacts; useful pattern source, not deploy-ready as-is |
| [ruview.md](reviews/ruview.md) | [ruvnet/RuView](https://github.com/ruvnet/RuView) | MIT | 📚 Study | Ambitious WiFi CSI spatial-sensing platform with Rust signal crates, ESP32 firmware, dashboards, Home Assistant/Matter docs, and MCP/CLI wrappers; useful pattern source but not a turnkey safety/health deployment |
| [openhuman.md](reviews/openhuman.md) | [tinyhumansai/openhuman](https://github.com/tinyhumansai/openhuman) | GPL-3.0 | ⚠️ Interesting | Personal AI desktop harness with Rust/Tauri core, React app, local memory tree, Markdown vault, integrations, messaging channels, MCP surfaces, and significant privacy/security caveats |
| [codegraph.md](reviews/codegraph.md) | [colbymchenry/codegraph](https://github.com/colbymchenry/codegraph) | MIT | ✅ Deploy candidate | Local pre-indexed code graph for coding agents with bundled runtime installs, SQLite/FTS5, tree-sitter plus optional native extraction kernel, single-default `codegraph_explore` MCP surface, broad framework dispatch coverage, and telemetry/dependency caveats |
| [cloudsail.md](reviews/cloudsail.md) | [nkzw-tech/cloudsail](https://github.com/nkzw-tech/cloudsail) | MIT | ✅ Deploy candidate | Self-hosted Cloudflare coding-agent sandboxes with Worker-owned credential injection, controlled egress, terminal/dev-server previews, git helpers, and clean alpha verification |
| [inbox-zero.md](reviews/inbox-zero.md) | [elie222/inbox-zero](https://github.com/elie222/inbox-zero) | AGPL-3.0 + commercial/enterprise restrictions | ⚠️ Interesting | Full AI email assistant with Gmail/Outlook automation, confirmed action gates, encrypted provider tokens, self-hosting docs, and dependency/license caveats |
| [tldraw.md](reviews/tldraw.md) | [tldraw/tldraw](https://github.com/tldraw/tldraw) | tldraw source-available + MIT templates | ✅ Deploy candidate | Infinite-canvas React SDK and whiteboard platform with multiplayer sync, AI starter kits, MCP app, and strong canvas-agent action patterns |
| [twenty.md](reviews/twenty.md) | [twentyhq/twenty](https://github.com/twentyhq/twenty) | AGPL-3.0 + Enterprise-marked files | ✅ Deploy candidate | Open-source CRM/product platform with self-hosted Docker/Kubernetes deployment, NestJS/React monorepo, SDK/app scaffolder, custom auth lint rules, and Claude skill direction |
| [anytype-ts.md](reviews/anytype-ts.md) | [anyproto/anytype-ts](https://github.com/anyproto/anytype-ts) | ASAL-1.0 source-available | 📚 Study | Official Anytype Electron/React desktop client for local-first encrypted object graphs, spaces, blocks, relations, sync middleware, and extension bridging |
| [bumblebee.md](reviews/bumblebee.md) | [perplexityai/bumblebee](https://github.com/perplexityai/bumblebee) | Apache-2.0 | ✅ Deploy candidate | Read-only developer endpoint scanner for exact package/tool exposure checks across lockfiles, package metadata, extensions, MCP configs, and threat-intel catalogs |
| [claude-ads.md](reviews/claude-ads.md) | [AgriciDaniel/claude-ads](https://github.com/AgriciDaniel/claude-ads) | MIT | ✅ Deploy candidate | Multi-platform paid-advertising audit skill pack with platform playbooks, helper agents, local scripts, PDF/report tooling, and a pytest eval harness |
| [multica.md](reviews/multica.md) | [multica-ai/multica](https://github.com/multica-ai/multica) | source-available Apache-2.0-derived | ✅ Deploy candidate | Managed coding-agent workspace with web, desktop, mobile, daemon, cloud/local runtimes, squads, autopilots, and task lifecycle control |
| [activegraph.md](reviews/activegraph.md) | [yoheinakajima/activegraph](https://github.com/yoheinakajima/activegraph) | Apache-2.0 | ⚠️ Interesting | Event-sourced reactive graph runtime for auditable, replayable, forkable agent systems |
| [text-to-cad.md](reviews/text-to-cad.md) | [earthtojake/text-to-cad](https://github.com/earthtojake/text-to-cad) | MIT | ✅ Deploy candidate | Agent skill bundle for source-first CAD, robotics descriptions, geometry inspection, render review, and CAD project harnessing |
| [cal-diy.md](reviews/cal-diy.md) | [calcom/cal.diy](https://github.com/calcom/cal.diy) | MIT | ⚠️ Interesting | MIT-only self-hosted Cal.com fork for personal/non-production scheduling infrastructure |
| [noah-zender-ideas.md](reviews/noah-zender-ideas.md) | [Noah Zender Ideas](https://www.noahzender.com/ideas) | N/A | 📚 Good reference | Large index of short mental-model notes for product, AI, writing, markets, psychology, and leadership |
| [agent-governance-toolkit.md](reviews/agent-governance-toolkit.md) | [microsoft/agent-governance-toolkit](https://github.com/microsoft/agent-governance-toolkit) | MIT | 📚 Study | Microsoft public-preview runtime governance stack for AI agents: policy, identity, MCP security, audit, SRE, and compliance |
| [freellmapi.md](reviews/freellmapi.md) | [tashfeenahmed/freellmapi](https://github.com/tashfeenahmed/freellmapi) | MIT | ⚠️ Interesting | Local OpenAI-compatible proxy that aggregates free-tier LLM providers with key-aware fallback routing |
| [gait.md](reviews/gait.md) | Clyra-AI/gait | Apache-2.0 | 🔥🔥🔥 | Policy-as-code enforcement at the AI agent tool boundary |
| [sage.md](reviews/sage.md) | l33tdawg/sage | Apache-2.0 | 🔥🔥🔥 | BFT consensus-validated persistent memory for AI agents |
| [crucix.md](reviews/crucix.md) | calesthio/Crucix | AGPL-3.0 | 🔥🔥🔥🔥 | Self-hosted OSINT intelligence terminal, 27 sources |
| [flash-moe.md](reviews/flash-moe.md) | danveloper/flash-moe | **no license** | 🔥🔥🔥🔥🔥 | Pure C/Metal inference for 397B MoE on a 48GB MacBook |
| [claude-chromium-native-messaging.md](reviews/claude-chromium-native-messaging.md) | stolot0mt0m/claude-chromium-native-messaging | MIT | 🔥🔥 | Claude extension for non-Chrome Chromium browsers |
| [openalice.md](reviews/openalice.md) | TraderAlice/OpenAlice | AGPL-3.0 | 🔥🔥🔥🔥 | File-driven AI trading agent engine — UTA pattern, Trading-as-Git, guard pipeline, evolution mode |
| [the-library.md](reviews/the-library.md) | disler/the-library | MIT | 🔥🔥🔥🔥 | Meta-skill for private-first distribution of skills/agents/prompts across devices and teams |
| [supabase.md](reviews/supabase.md) | supabase/supabase | Apache-2.0 | ⭐⭐⭐⭐⭐ | Open-source Firebase alternative — Postgres, auth, realtime, edge functions, self-hostable |
| [citadel.md](reviews/citadel.md) | SethGammon/Citadel | MIT | 📚 Study | Agent orchestration OS for Claude Code — proportional routing, campaign persistence, fleet coordination |
| [ctx.md](reviews/ctx.md) | stevesolun/ctx | MIT | 📚 Study | Graph-backed recommendations for capped skill, agent, MCP, and harness context bundles |
| [decapod.md](reviews/decapod.md) | DecapodLabs/decapod | MIT | 📚 Study | Repo-local governance kernel for AI coding agents — intent, context, boundaries, proof |
| [nobulex.md](reviews/nobulex.md) | arian-gogani/nobulex | MIT | 📚 Study | Proof-of-behavior protocol for earned agent autonomy and Trust Capital |
| [photogimp.md](reviews/photogimp.md) | Diolinux/PhotoGIMP | GPL-3.0 | ✅ Deploy candidate | GIMP 3.0 profile patch that gives Photoshop users familiar shortcuts, layout, icons, and splash assets |
| [hermes-agent-control-room.md](reviews/hermes-agent-control-room.md) | shannhk/hermes-agent-control-room | MIT | 🔧 Harvest | Sidecar control room for VPS-hosted agent teams, runbooks, secret maps, task bus, and SOP skills |
| [ruflo.md](reviews/ruflo.md) | ruvnet/ruflo | MIT | 📚 Study | Claude Code orchestration platform with plugins, swarms, memory, federation, and signed verification witnesses |
| [vibe-coding.md](reviews/vibe-coding.md) | wanderloots-tutorials/vibe-coding | **no license** | 📚 Study | Tutorial notes with an agent-buildable Raw/Wiki/Schema knowledge workflow |
| [motus.md](reviews/motus.md) | lithos-ai/motus | Apache-2.0 | ⚠️ Interesting | Python agent serving/runtime framework with session APIs, task graph workflows, MCP/tools, and cloud deploy |
| [local-notebooklm.md](reviews/local-notebooklm.md) | Goekdeniz-Guelmez/Local-NotebookLM | Apache-2.0 | ⚠️ Interesting | Local NotebookLM-style PDF-to-audio app with CLI, FastAPI, Gradio, and Docker surfaces |
| [insforge.md](reviews/insforge.md) | [InsForge/InsForge](https://github.com/InsForge/InsForge) | Apache-2.0 | ⚠️ Interesting | Agent-native backend platform with Postgres, auth, storage, functions, MCP/CLI surfaces, and typed control-plane APIs |
| [audiblez.md](reviews/audiblez.md) | [santinic/audiblez](https://github.com/santinic/audiblez) | MIT | ⚠️ Interesting | Local EPUB-to-M4B audiobook generator using Kokoro TTS, chapter WAV intermediates, and ffmpeg packaging |
| [stainful.md](reviews/stainful.md) | [stainlu/stainful](https://github.com/stainlu/stainful) | MIT | 📚 Study | Local Stainless-compatible Python SDK generator with an IR-first pipeline and conformance-gated output |
| [deepsec.md](reviews/deepsec.md) | [vercel-labs/deepsec](https://github.com/vercel-labs/deepsec) | Apache-2.0 | ✅ Deploy candidate | Agent-powered vulnerability scanner with resumable per-file records, PR mode, revalidation, plugins, and sandbox workers |
| [codebase-reasoning-topology-gist.md](reviews/codebase-reasoning-topology-gist.md) | [acidgreenservers gist](https://gist.github.com/acidgreenservers/001185d63e5cd65f9fbe6f7a1c70a200) | **no license** | 📚 Good reference | Coding-agent prompt bundle for topology-first codebase reasoning and read-only reconnaissance |
| [9router.md](reviews/9router.md) | [decolua/9router](https://github.com/decolua/9router) | MIT | ⚠️ Interesting | Local AI router/dashboard for coding tools with provider fallback, quota tracking, translation, MITM/tunnel helpers, and RTK token compression |
| [agentmemory.md](reviews/agentmemory.md) | [rohitg00/agentmemory](https://github.com/rohitg00/agentmemory) | Apache-2.0 | ⚠️ Interesting | Persistent memory server for AI coding agents with hook capture, MCP/REST, hybrid retrieval, graph/lessons, commit provenance, privacy, audit, and retention |
| [supertonic.md](reviews/supertonic.md) | [supertone-inc/supertonic](https://github.com/supertone-inc/supertonic) | MIT code / OpenRAIL-M model | ✅ Deploy candidate | On-device multilingual TTS reference kit with ONNX Runtime examples across Python, browser/WebGPU, Node, Rust, Swift, Go, Java, C++, C#, Flutter, and iOS |
| [cloakbrowser.md](reviews/cloakbrowser.md) | [CloakHQ/CloakBrowser](https://github.com/CloakHQ/CloakBrowser) | MIT wrapper / proprietary binary | ⚠️ Interesting | Security-sensitive Playwright/Puppeteer wrapper around a patched Chromium binary, with strong wrapper tests, binary provenance docs, and anti-detection risk |
| [html-anything.md](reviews/html-anything.md) | [nexu-io/html-anything](https://github.com/nexu-io/html-anything) | Apache-2.0 | ⚠️ Interesting | Local-first agentic HTML editor that drives coding-agent CLIs through skill templates, streaming preview, and platform-specific exports |

---

## Agents

| File | Source | Description |
|------|--------|-------------|
| [durable-object-workspace-runtime.md](patterns/durable-object-workspace-runtime.md) | cloudflare/computer | Keep a platform-native durable workspace as the source of truth, then route execution to pluggable lightweight isolate or heavyweight container backends with explicit capability descriptions and sync boundaries |
| [autoforge-two-agent-pattern.md](agents/autoforge-two-agent-pattern.md) | leonvanzyl/autoforge | Two-agent architecture for autonomous coding (initializer + coder) |
| [shannon-ai-pentester.md](agents/shannon-ai-pentester.md) | keygraphHQ/shannon | Autonomous AI pentester with Temporal orchestration |
| [antfarm-multi-agent-workflows.md](agents/antfarm-multi-agent-workflows.md) | snarktank/antfarm | Multi-agent workflows with YAML + SQLite + cron |
| [raptor-security-research-framework.md](agents/raptor-security-research-framework.md) | gadievron/raptor | Security research framework with progressive loading |
| [claude-code-controller-orchestration.md](agents/claude-code-controller-orchestration.md) | pacholoamit/claude-code-controller | Orchestrate real Claude Code processes via REST/SDK |
| [koda-architecture.md](agents/koda-architecture.md) | koda | Subagent spawning architecture |
| [monty-sandboxed-interpreter.md](agents/monty-sandboxed-interpreter.md) | monty | Sandboxed Python interpreter pattern |
| [ecc-patterns/](agents/ecc-patterns/) | affaan-m/everything-claude-code (MIT) | ECC instincts, hooks, orchestration patterns |

## Patterns

| File | Source | Description |
|------|--------|-------------|
| [static-first-semantic-diagram-skill.md](patterns/static-first-semantic-diagram-skill.md) | cathrynlavery/diagram-design | Generate diagrams by choosing semantic behavior patterns before visual layout, rendering a complete static HTML/SVG frame by default, and allowing motion only through a reviewed controller plus deterministic self-checks |
| [llm-edit-plan-compiler-boundary.md](patterns/llm-edit-plan-compiler-boundary.md) | boldsoftware/meat | Let an LLM propose structured operations over stable source coordinates while deterministic local code validates and renders the final transformed artifact, keeping model judgment useful without trusting model-authored output |
| [detached-eval-control-channel.md](patterns/detached-eval-control-channel.md) | Inspect AI | Launch long-running evals as supervised detached jobs with a machine-readable startup handoff, local-only control channel, live task/sample/transcript inspection, pause/resume/cancel/config controls, and explicit completion records |
| [agent-evals-as-test-files.md](patterns/agent-evals-as-test-files.md) | OpenRouter Ori Eval | Represent agent/model evals as ordinary `*.eval.ts` test files while a CLI owns runtime startup, SDK injection, credential resolution, paid model execution, append-only result capture, reports, and history |
| [browser-only-byok-media-generation.md](patterns/browser-only-byok-media-generation.md) | elder-plinius/GL4SS | Build static AI media apps where the user's provider key stays in the browser, authenticated calls are CSP/origin bounded, paid generation is explicit, generated artifacts persist locally, and demand/prefetch work is separately controlled |
| [lossless-off-wire-context-governor.md](patterns/lossless-off-wire-context-governor.md) | gbgh1/context-governor | Keep the model-facing prompt as a bounded projection over a fuller local context store, replacing bulky content with stable handles and rehydrating exact content under budget when needed |
| [profile-backed-agent-roster.md](patterns/profile-backed-agent-roster.md) | NousResearch/Hermes-Bot-Mode | Represent named bots as UI projections over existing runtime profiles, with canonical chat pins, scoped routines, metadata overlays, and agent-readable handoff protocol instead of a second bot database |
| [frame-locked-scroll-video-world.md](patterns/frame-locked-scroll-video-world.md) | oso95/scroll-world | Build scroll-scrubbed cinematic pages by chaining AI-generated videos through connectors anchored on the actual rendered endpoint frames, with blob-backed seeking, coalesced scroll updates, reduced-motion fallback, budget gates, and native mobile composition |
| [trajectory-aware-model-routing.md](patterns/trajectory-aware-model-routing.md) | codejunkie99/sageroute | Route agent sessions by observed tool-call evidence instead of prompt-time difficulty guesses: recover trajectory, compute deterministic stuck/progress signals, ask a decision policy only at checkpoints, then continue, switch model, restart clean, or escalate human with auditable headers |
| [file-output-research-agent-benchmark.md](patterns/file-output-research-agent-benchmark.md) | perplexityai/wandr | Evaluate research agents through task-declared required files, provider-specific artifact collection, workspace-relative validation, verifier-owned fetching/entity/evidence scoring, and trace-complete observability |
| [simd-pretokenization-cache-pipeline.md](patterns/simd-pretokenization-cache-pipeline.md) | marcelroed/gigatoken | Treat tokenizer throughput as a whole native data path with file loading, boundary-safe chunking, SIMD pretokenization, repeated-pretoken caching, cache-friendly BPE lookup, and ragged output buffers |
| [sealed-agent-trajectory-judge.md](patterns/sealed-agent-trajectory-judge.md) | cosmtrek/mindwalk | Evaluate untrusted agent-session traces with a sealed no-tool/no-MCP/no-config judge subprocess, evidence-cited JSON findings, deterministic verdict rollup, and stale-report detection |
| [diagnosis-before-cure-intake.md](patterns/diagnosis-before-cure-intake.md) | Corey Ganim AI Tools Assessment thread | Start with a transcript-first diagnostic conversation about lived workflow friction before recommending tools, process changes, automation, or support |
| [git-derived-agent-worker-scope-gate.md](patterns/git-derived-agent-worker-scope-gate.md) | procoders/superpowers-v | Verify each coding-agent worker's actual changed-file set from git against manifest-declared write scopes before accepting or merging the result |
| [deterministic-tool-call-benchmark-harness.md](patterns/deterministic-tool-call-benchmark-harness.md) | SeraphimSerapis/tool-eval-bench | Evaluate model/tool-serving quality with stable deterministic scenarios, mock tools, evaluator-local expected behavior, safety-gated scoring, endpoint adapters, and trace-complete reports |
| [closed-loop-analytics-skill-readback.md](patterns/closed-loop-analytics-skill-readback.md) | ericosiu/ai-marketing-skills | Promote skill, prompt, rubric, or playbook changes only after baseline/candidate outcome readback proves they worked |
| [root-mediated-model-routing-policy.md](patterns/root-mediated-model-routing-policy.md) | Cjbuilds/Codex-Orchestration | Keep one root agent authoritative while optional Planner, Advisor, and Executor seats provide bounded planning, critique, and implementation routes |
| [guardrails-not-choreography-plans.md](patterns/guardrails-not-choreography-plans.md) | EveryInc/compound-engineering-plugin | Use plans as durable decision artifacts with scope, stable unit IDs, verification scenarios, and risks, while leaving exact code shape and execution steps to the implementer with current context |
| [disk-tiered-moe-inference.md](patterns/disk-tiered-moe-inference.md) | JustVugg/colibri | Run oversized MoE models by mapping routed experts across NVMe, RAM, and VRAM tiers, using route history, resource planning, multi-drive layouts, and hardware-specific accelerator residency while preserving model semantics |
| [secretless-sandbox-credential-proxy.md](patterns/secretless-sandbox-credential-proxy.md) | omnigent-ai/omnigent | Keep real credentials in a trusted parent process and let an egress proxy attach or swap them only for configured hosts, so sandboxed agent tools get scoped API access without receiving long-lived tokens |
| [file-backed-subagent-review-loop.md](patterns/file-backed-subagent-review-loop.md) | obra/superpowers | Use workspace artifacts for task briefs, diffs, implementer reports, reviewer verdicts, and progress ledgers so subagent-driven development stays resumable, auditable, and less context-heavy |
| [scanner-backed-aesthetic-taxonomy-skill.md](patterns/scanner-backed-aesthetic-taxonomy-skill.md) | yetone/kill-ai-slop | Turn subjective design critique into an agent workflow with a named taxonomy, detection leads, false-positive notes, remediation patterns, a read-only scanner, human-visible triage, and approval-gated minimal fixes |
| [tiered-kv-cache-inference-layer.md](patterns/tiered-kv-cache-inference-layer.md) | LMCache/LMCache | Treat LLM KV cache as an explicit serving layer with engine connectors, token/chunk indexes, GPU/CPU transfer adapters, L1/L2 storage tiers, cache-control APIs, and hit/transfer observability |
| [privacy-bounded-personal-text-model-pipeline.md](patterns/privacy-bounded-personal-text-model-pipeline.md) | Doriandarko/texts-to-transformer | Train or evaluate local models on private communications through read-only snapshots, pseudonymization, redaction, chronological guarded splits, train-only tokenizer, aggregate memorization checks, sensitive artifact handling, and no-send generation |
| [motion-review-quality-gate.md](patterns/motion-review-quality-gate.md) | emilkowalski/skills | Review UI motion as a first-class quality surface with explicit blockers for high-frequency animation, weak easing, wrong origins, layout-property animation, missing reduced-motion handling, and vague taste-only feedback |
| [anti-convergence-deliberation-protocol.md](patterns/anti-convergence-deliberation-protocol.md) | 0xNyk/council-of-high-intelligence | Run hard decisions through blind-first analysis, anonymized cross-examination, named updates, forced dissent checks, structured stance lines, confidence-weighted tallies, no-consensus reporting, and separate synthesis |
| [git-blob-keyed-code-memory.md](patterns/git-blob-keyed-code-memory.md) | farhankhwaja/codesherpa | Cache expensive code-index work by git blob hash, map active refs/paths separately, recompute path-dependent projections, and validate incremental sync against clean rebuilds |
| [living-slice-graph-spec-loop.md](patterns/living-slice-graph-spec-loop.md) | dzhng/skills | Represent long-running software work as a living spec folder with independently verifiable slices, current handoff state, review evidence, gates, and explicit reslicing when implementation uncovers hidden scope |
| [per-method-versioned-rpc-compatibility.md](patterns/per-method-versioned-rpc-compatibility.md) | traycerai/traycer | Evolve independently shipped clients, CLIs, and local hosts with per-method schema versions, manifest negotiation, explicit upgrade/downgrade paths, and transport-owned payload transforms |
| [oauth-scoped-application-mcp-tools.md](patterns/oauth-scoped-application-mcp-tools.md) | outline/outline | Expose an application's native objects/actions as MCP tools by creating a scope-filtered per-request tool surface, gating it by tenant preference, and reusing existing auth, policy checks, presenters, signed URLs, rate limits, and tests |
| [edge-agent-command-allowlist.md](patterns/edge-agent-command-allowlist.md) | getarcaneapp/arcane | Restrict remote high-authority agents through an explicit method/path/stream command registry, stable command names, manager-side permission checks, audit labels, and fail-closed routing |
| [cli-subagent-context-offload.md](patterns/cli-subagent-context-offload.md) | Use Codex Skill | Delegate noisy or context-heavy work to bounded CLI subagents while the parent agent owns prompt specificity, side-effect boundaries, completion monitoring, claim validation, and final synthesis |
| [cve-to-verified-container-lab-pipeline.md](patterns/cve-to-verified-container-lab-pipeline.md) | gadievron/cve-env | Build vulnerable lab environments through staged CVE research, image/source acquisition, hardened local launch, affected-version proof, benign functional smoke, and explicit partial outcome statuses |
| [llm-exploitability-validation-prompt-pipeline.md](patterns/llm-exploitability-validation-prompt-pipeline.md) | gadievron/exploitation-validator | Split LLM-driven vulnerability analysis into inventory, attempted proof, systematic attack-path exploration, source sanity checks, ruling filters, and final self-review artifacts |
| [public-bind-http-client-guard.md](patterns/public-bind-http-client-guard.md) | opendatalab/MinerU | Disable caller-supplied remote HTTP endpoints automatically when a self-hosted API is bound to public interfaces, unless the operator explicitly opts in |
| [advisor-executor-plan-handoff.md](patterns/advisor-executor-plan-handoff.md) | shadcn/improve | Separate senior codebase audit and planning from bounded execution by writing commit-stamped, self-contained plans with drift checks, scope, verification gates, and STOP conditions |
| [local-first-read-only-cloud-audit.md](patterns/local-first-read-only-cloud-audit.md) | azz-kikkr/kulshan | Run cloud posture and cost audits from a local CLI with explicit read-only IAM, opt-in packs, local reports, PII redaction, machine-readable output, and no SaaS data upload |
| [conformance-first-service-emulator.md](patterns/conformance-first-service-emulator.md) | nubo-db/dynoxide | Build local cloud-service emulators around real-service conformance suites, embedded test APIs, wire-compatible servers, explicit unsupported surfaces, and release gates for compatibility/security drift |
| [agent-native-drawio-diagram-pipeline.md](patterns/agent-native-drawio-diagram-pipeline.md) | Agents365-ai/drawio-skill | Generate editable draw.io artifacts through a pipeline of diagram planning, XML generation, structural linting, preview export, targeted XML repair, final embedded export, and exporter-specific repair |
| [agent-published-html-preview-pipeline.md](patterns/agent-published-html-preview-pipeline.md) | utsengar/htmlbin-cli | Publish agent-generated single-file HTML artifacts through a CLI that returns stable URLs and machine-readable results across public, GitHub Pages, or Cloudflare Pages backends |
| [local-loopback-agent-session-control.md](patterns/local-loopback-agent-session-control.md) | modem-dev/hunk | Give a human-facing local app a loopback-only structured control plane so agents can inspect, navigate, reload, and annotate live UI state without terminal scraping |
| [agent-minimalism-ladder.md](patterns/agent-minimalism-ladder.md) | DietrichGebert/ponytail | Force coding agents through skip, stdlib, native platform, existing dependency, and one-line rungs before custom code, while preserving validation, security, accessibility, and explicit requirements |
| [perspective-guided-research-pipeline.md](patterns/perspective-guided-research-pipeline.md) | stanford-oval/storm | Split long-form research generation into perspectives, simulated question asking, retrieval evidence tables, outline synthesis, citation-grounded drafting, and final polishing |
| [open-knowledge-format.md](patterns/open-knowledge-format.md) | GoogleCloudPlatform/knowledge-catalog | Represent operational knowledge as git-native markdown/frontmatter bundles with concept IDs, links, generated indexes, citations, and agent-safe preservation guards |
| [evidence-backed-command-output-compaction.md](patterns/evidence-backed-command-output-compaction.md) | nik1t7n/context-firewall | Run noisy commands through a local wrapper that stores exact stdout/stderr evidence, returns deterministic compact summaries, and exposes span handles for targeted raw retrieval |
| [direct-http-provider-parity-layer.md](patterns/direct-http-provider-parity-layer.md) | lfnovo/esperanto | Build a small direct-HTTP AI provider adapter layer around typed common responses, provider profiles, explicit unsupported-feature behavior, and separate mocked vs real-API tests |
| [deterministic-agent-workflow-graphs.md](patterns/deterministic-agent-workflow-graphs.md) | fabro-sh/fabro | Represent long-running agent work as version-controlled workflow graphs with typed nodes, explicit human gates, sandbox boundaries, model routing, durable events, and checkpoints |
| [canonical-agent-ops-repo.md](patterns/canonical-agent-ops-repo.md) | steipete/agent-scripts | Keep shared coding-agent instructions, skills, docs, helper scripts, and validators in one versioned repository, with downstream projects pointing at the canonical source instead of copying it |
| [scanner-safe-community-agent-skills.md](patterns/scanner-safe-community-agent-skills.md) | tmchow/agent-skills | Design community-installable skills to avoid scanner red flags: no ambient secret reads, no CLI secret flags, user-run config init, pinned installs, lean bundles, runtime metadata, and checksum-repaired binary assets |
| [persona-voice-safety-switch.md](patterns/persona-voice-safety-switch.md) | Lagunaswift/RockyVoice | Keep strong assistant personas persistent but subordinate to explicit off-switches, exact technical content, plain-language safety overrides, and separate local voice side channels |
| [stateless-baseline-gain-evaluation.md](patterns/stateless-baseline-gain-evaluation.md) | pgasawa/continual-learning-bench | Evaluate stateful agents by running the same canonical instances with memory enabled and reset between instances, then reporting reward, cost, and latency gain over the reset baseline |
| [local-git-remote-validation-gate.md](patterns/local-git-remote-validation-gate.md) | kunchenguid/no-mistakes | Put a local git remote and daemon-backed post-receive pipeline in front of upstream delivery so pushed branches are validated in disposable worktrees before PR/CI completion |
| [source-model-to-runtime-artifact-pipeline.md](patterns/source-model-to-runtime-artifact-pipeline.md) | trymirai/lalamo | Treat model conversion as a compiler pipeline from typed source-model specs through config translation, weight loading, compression, stable artifact export, and tiered compatibility validation |
| [multi-language-rust-core-sdk.md](patterns/multi-language-rust-core-sdk.md) | trymirai/uzu | Keep performance-sensitive runtime logic in one Rust core, annotate domain APIs once, and generate Python, Node, Swift, Rust, and WASM SDK surfaces from the same contract |
| [structured-markdown-query-language.md](patterns/structured-markdown-query-language.md) | harehare/mq | Treat Markdown as a structured document tree with selectors, query/update functions, formatter/checker tooling, multi-format output, and agent-facing recipes instead of relying on brittle regex |
| [hook-gated-agent-delivery-pipeline.md](patterns/hook-gated-agent-delivery-pipeline.md) | 3awny/qship | Move long-running agent completion gates out of prompt prose and into filesystem artifacts, pre-tool hooks, stop hooks, freshness checks, persistence loops, and watchdogs |
| [falsification-driven-agent-coding-loop.md](patterns/falsification-driven-agent-coding-loop.md) | dsweet99/agent_coding | Run coding agents through plan, implement, fresh-context review, falsification of concerns, repair, bounded retry, exact acceptance, and durable artifacts |
| [social-signal-recency-research-engine.md](patterns/social-signal-recency-research-engine.md) | mvanhorn/last30days-skill | Research recent topics by fanning out across social, market, developer, video, and web sources, normalizing evidence, scoring engagement/freshness/relevance, clustering, and rendering a brief plus raw artifact |
| [source-native-heterogeneous-retrieval.md](patterns/source-native-heterogeneous-retrieval.md) | JinheonBaek/OmniRetrieval | Route queries to backend-specific text, SQL, SPARQL, or Cypher execution instead of flattening heterogeneous sources into one vector index |
| [workflow-backed-agent-skill-library.md](patterns/workflow-backed-agent-skill-library.md) | ragnar-pwninskjold/tech-snacks | Package agent procedures as routeable skills backed by references, templates, specialized agents, and workflow scripts for stateful orchestration |
| [sandboxed-execution-verified-security-agent-loop.md](patterns/sandboxed-execution-verified-security-agent-loop.md) | anthropics/defending-code-reference-harness | Run high-risk security agents in isolated workers, require concrete artifacts, verify those artifacts in fresh containers, and gate patches with executable build, reproducer, test, and re-attack checks |
| [provider-normalized-agent-walkthrough-pipeline.md](patterns/provider-normalized-agent-walkthrough-pipeline.md) | austeane/walkthrough | Convert heterogeneous agent session histories into a normalized evidence model, reduce noise deterministically, then generate source-backed walkthrough artifacts |
| [ide-wired-agent-harness.md](patterns/ide-wired-agent-harness.md) | can1357/oh-my-pi | Build coding agents around semantic IDE/runtime tools: LSP, DAP, snapshot-anchored edits, AST previews, structured subagents, resource URLs, mounted tool devices, and heuristic local memory |
| [edge-native-agentic-email-runtime.md](patterns/edge-native-agentic-email-runtime.md) | cloudflare/agentic-inbox | Build an AI email client at the edge with Email Routing inbound, Email Service outbound, per-mailbox Durable Objects, R2 attachments, Access JWTs, draft-first AI tools, and MCP integration |
| [formula-transparent-financial-analytics.md](patterns/formula-transparent-financial-analytics.md) | JerBouma/FinanceToolkit | Keep provider ingestion, statement normalization, financial formulas, controller APIs, and snapshot tests as separate layers so ratios, risk metrics, options, and valuation outputs remain inspectable |
| [ledger-backed-compaction-memory.md](patterns/ledger-backed-compaction-memory.md) | elpapi42/pi-observational-memory | Preserve long-session continuity by appending source-backed observations, durable reflections, and drop tombstones, then folding that ledger deterministically during compaction |
| [portable-agent-brain-adapters.md](patterns/portable-agent-brain-adapters.md) | codejunkie99/agentic-stack | Keep agent memory, skills, protocols, tools, and local telemetry in a project-local brain directory, then connect many coding-agent harnesses through manifest-validated adapters and conservative install/remove state |
| [local-document-vlm-runtime.md](patterns/local-document-vlm-runtime.md) | datalab-to/surya | Run document OCR/layout/table recognition through task-specific predictors over a shared OpenAI-compatible local VLM server with health probes, locks, sentinels, backend spawn paths, and structured outputs |
| [local-model-release-matrix.md](patterns/local-model-release-matrix.md) | openbmb/MiniCPM5-1B | Package local models as runtime-ready families with base/SFT/final checkpoints, GGUF/MLX variants, chat templates, deployment recipes, fine-tuning guides, and tool-call parser notes |
| [pre-install-agent-skill-security-scanner.md](patterns/pre-install-agent-skill-security-scanner.md) | NVIDIA/SkillSpector | Scan agent skills before installation with file inventory, manifest/parameter parsing, deterministic analyzer fan-out, optional semantic filtering, OSV-backed dependency checks, SARIF/JSON reports, and explicit allow/caution/deny policy decisions |
| [agent-memory-gateway.md](patterns/agent-memory-gateway.md) | supermemoryai/supermemory | Expose durable agent memory through scoped projects/container tags, MCP tools/resources/prompts, SDK wrappers, profile/search composition, and graph visualization while treating memory as an untrusted, privacy-sensitive tool surface |
| [zero-trust-agent-control-stack.md](patterns/zero-trust-agent-control-stack.md) | Zero Trust for AI Agents | Build agent deployments around cryptographic identity, short-lived credentials, least agency, tool-side validation, identity-based isolation, sandboxing, memory integrity, config integrity, traceability, and human-gated response |
| [tiered-blocklist-compiler-pipeline.md](patterns/tiered-blocklist-compiler-pipeline.md) | luisfelipess/ad-filter-list | Compile upstream feeds into tiered deployment artifacts with format-aware readers/writers, local overrides, safety guards, generated outputs, and provenance reports |
| [dependency-gated-tool-endpoints.md](patterns/dependency-gated-tool-endpoints.md) | Stirling-Tools/Stirling-PDF | Model optional local-tool capabilities as endpoint groups, dependency-backed tool groups, alternatives, disable reasons, startup probes, and central request enforcement |
| [local-first-agent-trace-store.md](patterns/local-first-agent-trace-store.md) | ssreeni1/tracebase | Capture local agent runs as encrypted raw blobs plus redacted searchable metadata, rebuildable SQLite/FTS projections, bounded dashboard/API/MCP views, and explicit raw-export gates |
| [lifecycle-managed-agent-skills.md](patterns/lifecycle-managed-agent-skills.md) | MUSE-Autoskill | Treat agent skills as versionable, testable, memory-bearing software assets with lifecycle operations for creation, validation, use, refinement, merge, and prune |
| [durable-knowledge-runtime-trace-bridge.md](patterns/durable-knowledge-runtime-trace-bridge.md) | apprentice-labs/activegraph-gbrain-bridge | Keep durable knowledge storage separate from runtime causality by bridging typed knowledge operations into redaction-aware events, evidence projections, context decisions, writeback proposals, forks, replay, and compatibility ledgers |
| [tolerance-routed-gpu-operator-dispatch.md](patterns/tolerance-routed-gpu-operator-dispatch.md) | FlashML-org/flashlib | Expose a numeric tolerance budget, route to the fastest acceptable GPU kernel/algorithm variant, and mirror runtime dispatch with a structured cost API for schedulers and agents |
| [rendered-page-ocr-export-pipeline.md](patterns/rendered-page-ocr-export-pipeline.md) | transitive-bullshit/kindle-ai-export | Use an authorized rendered reader as the source of truth, capture per-page images, OCR them, preserve navigation metadata, and export private derived formats with provenance |
| [code-as-action-browser-task-workspace.md](patterns/code-as-action-browser-task-workspace.md) | microsoft/Webwright | Make browser tasks produce a rerunnable script, per-run logs, screenshots, critical-point evidence, and verification gates instead of a hidden browser transcript |
| [live-worker-catalog-runtime.md](patterns/live-worker-catalog-runtime.md) | iii-hq/iii | Model backend capabilities as live-discoverable workers, functions, and triggers with permission-shaped discovery, unified invocation, and shared telemetry |
| [live-optimization-trace-viewer.md](patterns/live-optimization-trace-viewer.md) | modaic-ai/gepa-viz | Capture optimizer callbacks into replayable JSON traces, stream full snapshots to a live browser view, and inspect branch topology, prompt diffs, minibatch feedback, and per-example outcomes |
| [hardware-aware-local-ai-appliance-installer.md](patterns/hardware-aware-local-ai-appliance-installer.md) | Light-Heart-Labs/DreamServer | Detect host capability, choose models from a versioned catalog, generate secrets/env state, compose services via overlays/manifests, and enforce network exposure policy with release-contract tests |
| [12-factor-agent-architecture.md](patterns/12-factor-agent-architecture.md) | humanlayer/12-factor-agents | Design production agents as explicit software systems with owned prompts/context/state, structured outputs, human approval tools, compact errors, and reducer-style steps |
| [agentic-video-production-pipeline.md](patterns/agentic-video-production-pipeline.md) | HKUDS/ViMax | Turn a loose creative brief into durable story, character, storyboard, frame, clip, and final-video artifacts with provider-isolated media generation |
| [wifi-csi-edge-sensing-pipeline.md](patterns/wifi-csi-edge-sensing-pipeline.md) | ruvnet/RuView | Build a local WiFi CSI sensing pipeline with calibration, quality flags, confidence scores, bounded semantic outputs, and privacy-aware home-automation integration |
| [local-personal-ai-memory-harness.md](patterns/local-personal-ai-memory-harness.md) | tinyhumansai/openhuman | Build a local desktop assistant core around inspectable memory, Markdown vault export, credential isolation, prompt-injection gates, and explicit integration trust boundaries |
| [preindexed-code-intelligence-mcp.md](patterns/preindexed-code-intelligence-mcp.md) | colbymchenry/codegraph | Keep a local code graph beside a repository and expose one strong source-grounded MCP explore tool, with narrower search/context/callers/impact/status tools available when needed |
| [edge-brokered-coding-sandboxes.md](patterns/edge-brokered-coding-sandboxes.md) | nkzw-tech/cloudsail | Run agents in remote coding containers while an edge control plane owns credentials, egress policy, previews, lifecycle, checkpoints, and cost limits |
| [private-email-assistant-action-gates.md](patterns/private-email-assistant-action-gates.md) | elie222/inbox-zero | Constrain private mailbox agents with untrusted-content hardening, template-only generated fields, pending confirmations, account-bound execution, encrypted secrets, and audit context |
| [local-first-encrypted-object-graph-client.md](patterns/local-first-encrypted-object-graph-client.md) | anyproto/anytype-ts | Split a local-first knowledge app into a rich Electron/React object-graph client backed by encrypted/syncing middleware and event streams |
| [read-only-endpoint-exposure-scanner.md](patterns/read-only-endpoint-exposure-scanner.md) | perplexityai/bumblebee | Read bounded local package/tool metadata, match exact exposure catalogs, and emit stable current-state records without executing packages |
| [domain-skill-eval-harness.md](patterns/domain-skill-eval-harness.md) | AgriciDaniel/claude-ads | Test large agent skills with routing snapshots, check catalog coverage, deterministic scoring, and helper-script security regressions |
| [managed-agent-runtime-control-plane.md](patterns/managed-agent-runtime-control-plane.md) | multica-ai/multica | Separate agent work management from local/cloud runtime execution with capability reporting, task claiming, provider adapters, and realtime progress |
| [forkable-event-sourced-agent-runs.md](patterns/forkable-event-sourced-agent-runs.md) | yoheinakajima/activegraph | Persist agent work as an event-sourced graph so runs can be audited, replayed, forked, and structurally diffed |
| [agentic-cad-skill-workbench.md](patterns/agentic-cad-skill-workbench.md) | earthtojake/text-to-cad | Constrain agentic CAD with source-first generation, explicit targets, deterministic inspection, render handoff, and repair loops |
| [pre-execution-agent-policy-gate.md](patterns/pre-execution-agent-policy-gate.md) | microsoft/agent-governance-toolkit | Deterministic allow/deny/audit policy checks before agent tool calls, resource access, messages, or workflow steps execute |
| [skill-security-audit.md](patterns/skill-security-audit.md) | gadievron/security-check-skill | Security auditing pattern for AI skills |
| [knowledge-work-plugins.md](patterns/knowledge-work-plugins.md) | anthropics/knowledge-work-plugins | Role-specific plugin architecture |
| [graphiti-knowledge-graph.md](patterns/graphiti-knowledge-graph.md) | getzep/graphiti | Real-time knowledge graphs for agents |
| [x-research-skill.md](patterns/x-research-skill.md) | xBenJamminx/x-research-skill | X/Twitter research skill pattern |
| [explain-openclaw-architecture.md](patterns/explain-openclaw-architecture.md) | djmango/explain-openclaw | OpenClaw architecture guide |
| [tinyfish-web-agents.md](patterns/tinyfish-web-agents.md) | tinyfish-ai/tinyfish-cookbook | Web agents for browser automation |
| [inconvo-data-agents.md](patterns/inconvo-data-agents.md) | inconvoai/inconvo | Chat-with-data agents for databases |
| [focus-mode-with-urgency-detection.md](patterns/focus-mode-with-urgency-detection.md) | koda | Focus mode with urgency detection |
| [progressive-skill-loading.md](patterns/progressive-skill-loading.md) | koda | Load skills progressively |
| [resilience-utilities.md](patterns/resilience-utilities.md) | koda | Error handling patterns |
| [smart-history-trimming.md](patterns/smart-history-trimming.md) | koda | Context window management |
| [subconscious-innovation-loop.md](patterns/subconscious-innovation-loop.md) | koda | Background innovation pattern |
| [vouch-web-of-trust.md](patterns/vouch-web-of-trust.md) | vouch | Web of trust verification |
| [mcp-template-server/](patterns/mcp-template-server/) | coding-standards-mcp | MCP server template |
| [citadel-campaign-persistence.md](patterns/citadel-campaign-persistence.md) | SethGammon/Citadel | Multi-session state persistence via structured campaign markdown files |
| [graph-backed-context-bundle-selection.md](patterns/graph-backed-context-bundle-selection.md) | stevesolun/ctx | Select capped helper bundles from large skill/agent/MCP catalogs using graph-ranked evidence |
| [agent-governance-kernel.md](patterns/agent-governance-kernel.md) | DecapodLabs/decapod | Callable governance loop around AI agent inference |
| [earned-autonomy-gates.md](patterns/earned-autonomy-gates.md) | arian-gogani/nobulex | Graduated agent permissions based on verifiable compliant behavior history |
| [profile-overlay-distribution.md](patterns/profile-overlay-distribution.md) | Diolinux/PhotoGIMP | Ship curated app defaults as inspectable profile files instead of forking the host app |
| [sidecar-agent-control-room.md](patterns/sidecar-agent-control-room.md) | shannhk/hermes-agent-control-room | Sidecar operations control plane for multi-agent teams |
| [signed-regression-witnesses.md](patterns/signed-regression-witnesses.md) | ruvnet/ruflo | Signed marker manifests for keeping past bug fixes auditable across refactors |
| [source-to-compiled-wiki-gates.md](patterns/source-to-compiled-wiki-gates.md) | wanderloots-tutorials/vibe-coding | Raw source layer, compiled wiki notes, JSONL catalog, source coverage, and lint gates |
| [agent-operated-backend-control-plane.md](patterns/agent-operated-backend-control-plane.md) | InsForge/InsForge | Typed backend control plane for AI coding agents |
| [local-tts-audiobook-pipeline.md](patterns/local-tts-audiobook-pipeline.md) | santinic/audiblez | Resumable local TTS audiobook pipeline with chapter-level intermediates |
| [conformance-gated-sdk-codegen.md](patterns/conformance-gated-sdk-codegen.md) | stainlu/stainful | Test generated SDKs with committed output, regeneration stability, and oracle public-surface comparison |
| [sandbox-credential-brokering.md](patterns/sandbox-credential-brokering.md) | vercel-labs/deepsec | Keep real provider tokens on the orchestrator while sandboxed agent workers use egress-layer credential injection |
| [topology-first-codebase-reasoning.md](patterns/topology-first-codebase-reasoning.md) | acidgreenservers gist | Map state, feedback, blast radius, timing, and trust boundaries before non-trivial code changes |
| [rtk-tool-result-compression.md](patterns/rtk-tool-result-compression.md) | decolua/9router | Compress noisy coding-tool outputs before LLM routing while preserving errors and fallback safety |
| [hook-captured-agent-memory.md](patterns/hook-captured-agent-memory.md) | rohitg00/agentmemory | Build agent memory as a local service fed by lifecycle hooks, with privacy, progressive recall, provenance, audit, and retention |
| [on-device-onnx-tts-adapter.md](patterns/on-device-onnx-tts-adapter.md) | supertone-inc/supertonic | Wrap local ONNX TTS behind a small HTTP/OpenAI-compatible adapter so agents and apps can share one private speech runtime |
| [policy-gated-browser-automation-runtime.md](patterns/policy-gated-browser-automation-runtime.md) | CloakHQ/CloakBrowser | Put browser automation behind URL, launch-argument, network, binary-provenance, and audit policy gates |
| [agentic-html-surface-pipeline.md](patterns/agentic-html-surface-pipeline.md) | nexu-io/html-anything | Use local agent CLIs plus constrained skill templates, streaming iframe preview, and export adapters to generate finished HTML artifacts |
| [policy-gated-local-agent-credential-broker.md](patterns/policy-gated-local-agent-credential-broker.md) | asimons81/hermes-vault | Broker agent credential use through local encrypted storage, service/action policy, verification, TTL-limited env materialization, audit records, and raw-secret-safe operator surfaces |
| [evidence-bound-binary-critique-gate.md](patterns/evidence-bound-binary-critique-gate.md) | brandonsimpson/devils-advocate | Turn LLM critique into binary pass/fail criteria with context gates, file:line evidence, fix suggestions, independent review for self-critiques, and logged results |
| [scope-gated-security-skill-bundle.md](patterns/scope-gated-security-skill-bundle.md) | elementalsouls/Claude-BugHunter | Build high-risk domain skills around authorization gates, narrow routing, validation-before-action, evidence hygiene, and explicit operational stop-lines |
| [user-controlled-rag-research-workspace.md](patterns/user-controlled-rag-research-workspace.md) | lfnovo/open-notebook | Build research assistants around explicit notebooks, sources, notes, insights, context-inclusion levels, token budgets, and separate provider credentials |
| [agent-safe-social-api-cli.md](patterns/agent-safe-social-api-cli.md) | xdevplatform/xurl | Expose social APIs to agents through a local CLI with secret-free status checks, JSON shortcuts, forbidden secret flags/files, and approval gates for mutations |

## Prompts

| File | Source | Description |
|------|--------|-------------|
| [claudio-voice-personas.md](prompts/claudio-voice-personas.md) | cleanser-labs/claudio | Voice personas for TTS assistants |
| [chatgpt-prompts-library.md](prompts/chatgpt-prompts-library.md) | pacholoamit/chatgpt-prompts | 140+ curated GPT prompts |
| [shannon-pentesting-prompts.md](prompts/shannon-pentesting-prompts.md) | keygraphHQ/shannon | Pentesting agent prompts |
| [koda-soul-personality.md](prompts/koda-soul-personality.md) | koda | AI personality definition |
| [koda-conversation-summarizer.md](prompts/koda-conversation-summarizer.md) | koda | Conversation summarization |
| [koda-subagent-prompt.md](prompts/koda-subagent-prompt.md) | koda | Subagent task prompts |
| [supabase-coding-rules/](prompts/supabase-coding-rules/) | supabase/supabase (Apache-2.0) | 8 production coding rules — RLS, edge functions, auth, migrations, realtime |

## Tools

| File | Source | Description |
|------|--------|-------------|
| [skill-audit-cli.md](tools/skill-audit-cli.md) | markpors/skill-audit | CLI for auditing AI skills |
| [security-scanner-python.md](tools/security-scanner-python.md) | gadievron/security-check-skill | Python security scanner implementation |
| [langextract-structured-extraction.md](tools/langextract-structured-extraction.md) | langextract | Structured data extraction |
| [md-browse-turndown-config.md](tools/md-browse-turndown-config.md) | md-browse | HTML to Markdown conversion |
| [witr-process-causality.md](tools/witr-process-causality.md) | witr | Process causality tracking |

---

## Sources

All content includes attribution to its source repo, author, and license.

**Policy:** public repos → review published; license noted (including "none"); code only extracted under permissive licenses; AGPL/GPL summaries only; unlicensed → educational use only.

public-data

public-data 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

public-data

public-data 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README