{
  "markdown": "# OKF Atlas\n\n**A machine-readable knowledge layer for your repository, built for\ncoding agents.** One command turns your repo into a browsable index of\nevery file, symbol, dependency, and test — plus four LLM-drafted wiki\npages (architecture, entry points, core flows, change guide) that give\nagents and new contributors instant orientation. Bundles follow the\n[OKF spec](https://github.com/GoogleCloudPlatform/knowledge-catalog/blob/main/okf/SPEC.md).\n\n## Quick start\n\n```bash\ngit clone https://github.com/Fengzdadi/okf-atlas && cd okf-atlas\npnpm install && pnpm build\n```\n\n**Index your repo — no API key, no LLM, seconds:**\n\n```bash\npnpm exec okf-atlas generate /path/to/your/repo --stage1-only\n```\n\nOpen `/path/to/your/repo/docs/okf/index.md` and browse: per-file pages\nwith import graphs and reverse dependencies, one page per exported\nsymbol, per database table, per CLI entry point.\n\n**Add the wiki pages** (the part worth reading first) — no API key\nneeded if you have Claude Code or Codex installed. Just run\n`pnpm exec okf-atlas wiki /path/to/your/repo`: Atlas detects the\nsubscription CLIs on your machine and asks which one to draft through.\nTo script it instead:\n\n```bash\nexport OKF_ATLAS_LLM_PROVIDER=command\nexport OKF_ATLAS_LLM_COMMAND=\"$PWD/scripts/claude-code-provider.mjs\"   # or codex-provider.mjs\npnpm exec okf-atlas wiki /path/to/your/repo     # 4 calls, a few minutes\n```\n\n(Any OpenAI-compatible API key works too: `cp .env.example .env`, fill\nit in, `set -a; source .env; set +a`.)\n\n**Point agents at it** — one line in your repo's\n`CLAUDE.md`/`AGENTS.md`:\n\n> Consult `docs/okf/index.md` to navigate this repo.\n\nThat's it. Reruns are incremental and byte-deterministic; commit\n`docs/okf/` so your team and their agents get it for free.\n\n## What you get\n\n```\ndocs/okf/\n├── wiki/               # architecture, entrypoints, core-flows, change-guide\n├── modules/<path>.md   # per file: imports, exports, imported-by, tests\n├── apis/  data/  commands/  migrations/  external/  references/  tests/\n└── index.md            # everything, grouped by type\n```\n\n## Highlights\n\n- **Deterministic core** — everything except `wiki/` is extracted\n  statically (TS/JS via the compiler API, Python line-level, SQL,\n  OpenAPI, ADRs). Byte-identical reruns, CI-enforceable with\n  `generate --stage1-only && git diff --exit-code`.\n- **Cost safety** — `--estimate` projects LLM calls and payload before\n  you spend; `--max-calls` hard-caps; failures isolate per item and\n  retry next run; every run prints actual usage.\n- **Bring your API key or your subscription** — providers are\n  pluggable executables: OpenAI-compatible endpoints, Claude Code\n  (Claude plan), or Codex CLI (ChatGPT plan).\n- **Tunable per repo** — `.okf-atlas.yaml` scopes discovery, retypes\n  pages by glob, thresholds reference promotion, budgets the wiki.\n- **Tooling-friendly** — every command speaks `--format json`;\n  `lint --mode producer` validates any bundle in seconds.\n\nFull documentation: **[docs/guide.md](docs/guide.md)** ·\n[中文指南](docs/guide.zh-CN.md) · design contract:\n[SPEC-INTERNAL.md](SPEC-INTERNAL.md) · [ROADMAP.md](ROADMAP.md)\n\n## Status\n\nv0.1, pre-npm (`npx okf-atlas` coming). Validated end to end on\nexternal multi-package repositories; this repo indexes itself —\n`docs/okf/` here is Atlas's own output. MIT licensed. 中文说明见\n[README.zh-CN.md](README.zh-CN.md).\n",
  "bytes": 3358,
  "sha": "98c17006b8db8ceb4e76a8278280d8097ec20d47dc20aa313e5f6a80a1a09c55",
  "repo_slug": "fengzdadi/okf-atlas",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/okf_fengzdadi_okf_atlas_docs_okf_index_md_9b49fe58/readme"
}