codegraph-brain
AGENTS.md2 starsChanged 11 days agoMITView on GitHub
The guide for every AI agent in this repository. CLAUDE.md holds only @AGENTS.md, so Claude Code and every other agent read the same text. CodeGraph Brain (cgis — Code Graph Intelligence System) is a Python library and CLI that parses source code into a multi-layer semantic graph with fully qualified name (FQN) resolution. It uses tree-sitter for AST extraction and stores the resulting graph in SQLite. The CLI entry point is cgis ingest <path>, which produces a graph.json output. The…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository3 other files it gives its agents.
- ui/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- ui/CLAUDE.mdCLAUDE.md
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 0 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 77
Create src/cgis/languages/<name>/ with an extractor subclassing cgis.core.extractor.BaseExtractor (parse(code, file_path) -> …
- Very specific: line 88
Add one Manifest to MANIFESTS in src/cgis/languages/__init__.py, with a load function that imports the pack statically inside it (lazy, yet …
- Very specific: line 130
Read .entire/agent-guide.md for this repository's workflow, source inspection, and verification guidance.
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md — CodeGraph Brain
The guide for every AI agent in this repository. CLAUDE.md holds only @AGENTS.md,
so Claude Code and every other agent read the same text.
What This Is
CodeGraph Brain (cgis — Code Graph Intelligence System) is a Python library and CLI that parses source code into a multi-layer semantic graph with fully qualified name (FQN) resolution. It uses tree-sitter for AST extraction and stores the resulting graph in SQLite.
The CLI entry point is cgis ingest <path>, which produces a graph.json output.
Commands
# Run CLI
uv run cgis ingest <repo_path> --output graph.json
# Format
make format # runs: uv run ruff format .
# Lint (with auto-fix)
make lint # runs: uv run ruff check . --fix
# Type check (pyrefly against the committed error budget — this is mandatory)
make type-check # runs: uv run pyrefly check --baseline .github/pyrefly-baseline.json
# Tests
make pytest # runs: uv run pytest
# Docstring coverage (minimum 90%)
make doc-coverage # runs: uv run interrogate src
# Full verification (run before every commit/PR)
make format && make lint && make type-check && make pytest && make doc-coverage
# Single test
uv run pytest tests/unit/test_python_extractor.py::test_extract_simple_function -v
# Line coverage (same scope as CI — see the Makefile comment on why scripts/ counts)
make coverage # runs: uv run pytest --cov=cgis --cov=scripts --cov-report=term-missing
make coverage-html # same, writes htmlcov/Architecture
The pipeline runs in three phases: Extract → Resolve → Store.
IngestionPipeline (pipeline/__init__.py)
├── cgis.languages (registry) ← lazy: a pack is imported when its first file is met
│ ├── languages/python/ ← PythonExtractor + PACK
│ ├── languages/typescript/ ← TypeScriptExtractor (.ts/.tsx) + PACK
│ └── languages/go/ ← GoExtractor (.go) + go.mod reader + PACK (pilot)
├── ResolverEngine (resolver/engine.py) ← raw_call: targets → resolved FQN edges
├── SQLiteStore (storage/sqlite_store.py) ← persists graph; WAL mode
└── post-write passes (pipeline/) ← refresh_dispatch, SemanticUpliftEngine (pipeline/uplift.py)
└── QueryEngine (query/engine.py) ← BFS traversals (impact / flow)Node FQN format: module.ClassName.method_name — fully dot-separated, derived from the file path (e.g., src/cgis/pipeline/__init__.py → src.cgis.pipeline.IngestionPipeline.run). Use cgis.languages.scope_of(path, source_root) to convert paths (it picks the file's own pack; the Python rule lives in languages/python/extractor.py). __init__.py strips the /__init__ suffix.
Raw call convention: Extractors emit edges with target = "raw_call:<name>". The ResolverEngine then resolves these to actual FQNs, or leaves them unresolved: the target becomes the bare raw name (or a pack fallback such as js_builtins.<name>), a virtual boundary node classified STDLIB, EXTERNAL or UNKNOWN. A raw name equal to the id of a real node keeps the raw_call: prefix, so the edge never joins that node (#661).
ResolverEngine resolution order for a call to name:
self.method→ look upmethodon the enclosing class via_class_methodsindex- Unqualified call →
_global_symbolsby name, among the caller's own language only (#675); if ambiguous, prefer the same-scope candidate viafile_global_symbols(a file's scope is its path, or themetadata["scope"]its FILE node declares — a Go package)
Core Models (src/cgis/core/models.py)
Node— immutable (frozen Pydantic).idis the FQN. Key fields:type: NodeType,file_path,start_line,end_line,confidence_score,domains,ontology_class.Edge— immutable (frozen Pydantic).source/targetare FQNs. Key fields:type: EdgeType,weight,confidence. A call leaves the extractor as targetraw_call:<name>; after resolution a resolved call scores 1.0 in every language (#674) unless its target is an UNKNOWN boundary node (capped at 0.8, #493), and an unresolved one scores 0.8. A DISPATCHES edge scores 0.5, and an unresolved class reference 0.5.NodeType— structural (FILE,MODULE,CLASS,FUNCTION,METHOD…), runtime (API_ENDPOINT,DB_TABLE…), semantic (DOMAIN_CONCEPT).EdgeType— structural (CONTAINS,DECLARES,IMPORTS), behavioral (CALLS,REFERENCES), semantic (HANDLES,PERSISTS,AUTHORIZES).
Adding a Language Pack
- Create
src/cgis/languages/<name>/with an extractor subclassingcgis.core.extractor.BaseExtractor(parse(code, file_path) -> tuple[list[Node], list[Edge]]; emitCALLSedges withtarget="raw_call:<name>", setNode.language = "<name>"). - In its
__init__.py, definePACK, subclassingcgis.core.extractor.LanguagePack(mark methods@override) (name,extensions,code_fence,is_test,scope_of,extractor,project_reader,index_facts,classify,resolution).resolution()returnsResolutionRules(cgis.core.resolution): type normalizers, steps at a position (after:import_map), a fallback with its confidence,reconcile_import, and optionallydispatch(which methods a call through another type's may run; default: class hierarchy) — all reading the index asResolutionView. Metadata other files resolve against goes in the manifest'scross_file_metadata.classifyanswers for references from the pack's own files and runs the whole chain; a source with no pack gets the generic tail inSymbolIndex.classify_fqn. - Add one
ManifesttoMANIFESTSinsrc/cgis/languages/__init__.py, with aloadfunction that imports the pack statically inside it (lazy, yet visible to the review-fingerprint walk). Nothing else names the pack: CLI, MCP server, guardian and scripts all go through the registry, and thesurface-packsforbidden-edge gate fails if a surface imports a pack directly. - A pack never imports another pack or
cgis.storage(cross-pack,pack-isolationgates). - A pack's
__init__imports no grammar: import the extractor insideextractor(). Config reading goes in aProjectReader(cgis.core.project) whose facts are data keyed byingest_statekey — the collector (cgis.import_names), not the pack, writes them. A key other than the four built-ins (_FACT_SHAPES) travels asImportNames.extra: recorded, compared, and merged into thestatethe pack'sindex_factsreads. Its value must read back equal from JSON (no tuples). - A FILE node's id comes from
file_node_id, its definitions' namespace fromscope_of; they differ only where several files share a scope (Go), which the FILE node declares asmetadata["scope"].
Type Discipline
Type checking is pyrefly ([tool.pyrefly] in pyproject.toml approximates
the strict = true mypy configuration it replaces). All functions need full
annotations including return types; use model_copy(update={...}) for
immutable model updates.
The CI gate is an error budget: .github/pyrefly-baseline.json records the
individual type errors the tree has today, and the check fails when an error
appears that the baseline does not contain — it is a set, not a count, so
fixing one and adding another still fails. Fixing one means pruning the
baseline (uv run pyrefly check --baseline .github/pyrefly-baseline.json
--prune-baseline); code that moves means re-recording it with --update-baseline
on the same command. Never use --update-baseline to launder an error you just
introduced. The set does not depend on installed extras: the optional provider
SDKs are typed as Any in [tool.pyrefly], so a guardian-extra venv reports
the same errors CI does.
Test Layout
tests/unit/ ← pure unit tests, no I/O
tests/integration/ ← (empty, planned)
tests/self_parsing/ ← (empty) intended to validate the engine by parsing itselfThe self-parsing test (feeding src/cgis/ to the ingestion pipeline and asserting the resulting graph matches actual structure) is the canonical correctness validation described in the blueprint.
<!-- entire-agent:begin -->Read .entire/agent-guide.md for this repository's workflow, source inspection, and verification guidance. @.entire/agent-guide.md
<!-- entire-agent:end -->Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

