AGENTS.md
AGENTS.mdAGENTS.mdroot
Quality
89/100
Scores the file, not the repository.Length
411 words
8 headings · 1 code blocksRepository
4
— · pushed 27 days agoLast changed
3 days ago
First indexed 3 days ago.1# NSForge Agent Harness23Workspace instructions for autonomous agents (Copilot, Cline, Codex, OpenHands,4Hermes, …) working in the **Neurosymbolic Forge (NSForge)** repository.56## Goal78Help build and operate NSForge — an MCP server that turns *concepts* into9verifiable, traceable *entities* (symbols → derivations → algorithms). The AI10orchestrates; deterministic tools reify. See11`docs/reification-ladder-direction.md` for the north star.1213## North Star1415> Every symbol / equation / value / line of code in a final result must have a16> tool call as its "birth certificate" (provenance). The AI must not hand-derive17> any of them.1819Success = the amount the AI computes by hand approaches zero.2021## Working Style2223- Use Traditional Chinese unless the user asks otherwise.24- Prefer exact file paths, command output summaries, and verification results.25- Offload mechanical/deterministic steps (symbolic calculation, simplification,26 code generation) to tools — do not hand-derive formulas.27- When changing behavior, add a focused regression test under `tests/`.2829## Ground Truth: one command3031Before and after changes, verify against the green baseline:3233```bash34python scripts/check.py # all gates35python scripts/check.py --json # machine-readable summary (for agents)36python scripts/check.py --gates lint,type,test # subset37```3839Gates: lint, format, type, import, manifest, test, bench, generic, provenance, harness, diff. Exit code 0 = green.40(bench = derivation-correctness of `benchmarks/*.json` through the L3 orchestrator;41generic = arbitrary unseen compositions derive correctly, proving NSForge is a42derivation *calculus*, not a hand-built formula library; provenance = every43benchmark derivation carries a complete tool-provenance ledger, no hand-derived leaks;44harness = the harness self-checks its own invariants — version single-source,45self-describing tools, and gate/doc parity — so agents can trust the signal.)4647## Repo Layout (DDD)4849- `src/nsforge/domain/`: pure domain models/value objects (no I/O)50- `src/nsforge/application/`: use-cases / orchestration51- `src/nsforge/infrastructure/`: adapters (sympy engine, formula sources, file I/O)52- `src/nsforge_mcp/`: MCP tool layer + server wiring (`server.py`, `tools/`)53- `docs/agent/capabilities.json`: machine-readable tool manifest (self-describing)5455## Guardrails5657- Keep `src/nsforge/domain` free of I/O and infrastructure imports.58- Do not weaken `mypy` strict or skip gates just to make checks pass.59- Prefer `uv run ...` so tools use the project venv.60- Avoid destructive git ops (`reset --hard`, `clean -fdx`) unless asked.61- Never commit secrets or generated outputs (`dist/`, `.venv/`, `data/`).62- After adding/removing an `@mcp.tool`, regenerate the manifest:63 `python scripts/gen_capabilities.py`.6465## Related Files6667- `docs/reification-ladder-direction.md` — architecture direction68- `scripts/check.py` — verification harness (ground truth)69- `scripts/gen_capabilities.py` — capability manifest generator70- `.github/copilot-instructions.md` — Copilot rules71- `memory-bank/activeContext.md` — current focus72
Also in u9401066/nsforge-mcp
Diff this repo’s formatsOne repository carrying more than one format is the comparison this product exists for: does anyone actually write different content in each file, or is one a copy of the other?
| Repository | Format | Stack | Covers | Score | Changed |
|---|---|---|---|---|---|
| u9401066/nsforge-mcp.clinerules/00-project.md · 4 | Cline rules | testlint-formatstylearch+1 | 81/100 | 3 days ago | |
| u9401066/nsforge-mcp.clinerules/10-python.md · 4 | Cline rules | setuptestlint-formatstyle+1 | 83/100 | 3 days ago | |
| u9401066/nsforge-mcp.clinerules/40-release.md · 4 | Cline rules | lint-formatdeploymentdo-not | 51/100 | 3 days ago | |
| u9401066/nsforge-mcp.github/copilot-instructions.md · 4 | Copilot instructions | setuptestlint-formatgit+2 | 85/100 | 3 days ago | |
| u9401066/nsforge-mcpCLAUDE.md · 4 | CLAUDE.md | setuptestlint-formatgit+2 | 81/100 | 3 days ago |
Similar configs
Same format, overlapping stack, ranked by quality.
| Repository | Format | Stack | Covers | Score | Changed |
|---|---|---|---|---|---|
| vllm-project/vllmAGENTS.md · 88k | AGENTS.md | setuptestlint-formatstyle+5 | 100/100 | 3 days ago | |
| OnlyTerp/prompt-cache-skillsAGENTS.md · 112 | AGENTS.md | setupbuildtestlint-format+5 | 100/100 | 3 days ago | |
| n8n-io/n8npackages/@n8n/agents/AGENTS.md · 199k | AGENTS.md | buildteststylearch+3 | 100/100 | 3 days ago | |
| unoplat/unoplat-code-confluenceunoplat-code-confluence-frontend/AGENTS.md · 95 | AGENTS.md | setupbuildtestlint-format+6 | 100/100 | 2 days ago | |
| SkeneTechnologies/skene-cookbookAGENTS.md · 51 | AGENTS.md | setupbuildtestlint-format+7 | 100/100 | 2 days ago | |
| netdata/netdatasrc/go/plugin/ibm.d/AGENTS.md · 80k | AGENTS.md | buildtestlint-formatarch+3 | 99/100 | 3 days ago | |
| unoplat/unoplat-code-confluenceunoplat-code-confluence-query-engine/AGENTS.md · 95 | AGENTS.md | setupbuildtestlint-format+5 | 98/100 | 2 days ago | |
| alibaba/opc-starterAGENTS.md · 87 | AGENTS.md | setupbuildtestlint-format+5 | 97/100 | 3 days ago |
