learning-pi-through-force

KO | EN

gitlite — search

#javascript #python #hacktoberfest #typescript #llm #golang #android #machine-learning #deep-learning #linux #nodejs #windows #cli #mcp

learning-pi-through-force

(★ 12)

Source code for the learning pi through force blog

learning-pi-through-force 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# learning_pi_through_force

My [pi](https://github.com/) agent setup, packaged for sharing: the model
routing config for my self-hosted models, plus the custom **skills** and
**subagents** ("workflows") I've added on top of a stock pi install.

Everything here normally lives in `~/.pi/agent/`. Drop these files back into
that directory to reproduce the setup.

> **Note on secrets:** every credential has been replaced with a placeholder.
> Before use, fill in `models.json` (endpoint hosts + API keys) and copy
> `skills/hf-model-researcher/scripts/.env.example` → `.env` with real values.

## Layout

```
learning_pi_through_force/
├── models.json        # self-hosted model providers (endpoints + keys → placeholders)
├── settings.json      # subagent → model routing, thinking levels, default model
├── AGENTS.md          # working-style / delegation instructions
├── agents/            # custom subagents
│   ├── hf-retriever.md
│   └── prose.md
└── skills/            # custom skills
    ├── deep-research/
    ├── fetch-arxiv/
    ├── fetch-model/
    └── hf-model-researcher/
```

## Model config

**`models.json`** — three self-hosted OpenAI-compatible providers:

| Provider | Model | Role |
| --- | --- | --- |
| `lambda-glm` | GLM-5.2 | strong reasoning / orchestration |
| `lambda-kimi` | Kimi K2.7 Code | fast coder |
| `tailscale-deepseek` | DeepSeek V4 Flash | cheap scout / retrieval |

Replace `YOUR_ENDPOINT_HOST`, `YOUR_TAILSCALE_HOST`, and `YOUR_API_KEY` with
your own values.

**`settings.json`** — routes each subagent role to a model + thinking level
(e.g. `scout` → DeepSeek/off, `planner` → GLM-5.2/high, `worker` →
Kimi/medium with a GLM fallback). Default model is `lambda-glm/glm5.2`.

**`AGENTS.md`** — the working-style contract: delegate to the cheapest capable
specialist, run independent work in parallel, don't invent facts, don't leak
secrets.

## Subagents (workflows)

- **`hf-retriever`** — fast parallel source-cluster retriever for model
  research. Pinned to the fast Kimi model (GLM fallback) so the strong model is
  reserved for synthesis.
- **`prose`** — dedicated editorial agent for outlines, drafts, rewrites, and
  polished human-facing writing.

## Skills

- **`deep-research`** — enterprise-grade 8-phase research pipeline (Scope, Plan,
  Retrieve, Triangulate, Synthesize, Critique, Refine, Package) with citation
  tracking and verification. Includes Python scripts and report templates.
- **`fetch-arxiv`** — safely find/fetch arXiv papers via the structured API and
  `/html/` endpoint instead of guessing URLs.
- **`fetch-model`** — efficiently fetch HuggingFace model data (README,
  `config.json`, optional predecessor config) via parallel `curl`.
- **`hf-model-researcher`** — research a newly released HF model and produce a
  comprehensive report (model card + technical reports + novelty analysis +
  glossary + benchmarks), exported to markdown/PDF, optionally emailed or sent
  to Kindle. Email/Kindle require the `.env` (see `scripts/.env.example`).

## Install

```sh
# from a stock pi install
cp -r learning_pi_through_force/skills/*   ~/.pi/agent/skills/
cp -r learning_pi_through_force/agents/*   ~/.pi/agent/agents/
cp learning_pi_through_force/models.json   ~/.pi/agent/
cp learning_pi_through_force/settings.json ~/.pi/agent/
cp learning_pi_through_force/AGENTS.md     ~/.pi/agent/

# then fill in secrets
$EDITOR ~/.pi/agent/models.json
cp ~/.pi/agent/skills/hf-model-researcher/scripts/.env.example \
   ~/.pi/agent/skills/hf-model-researcher/scripts/.env
$EDITOR ~/.pi/agent/skills/hf-model-researcher/scripts/.env
```

learning-pi-through-force

learning-pi-through-force 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

learning-pi-through-force

learning-pi-through-force 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README