ci: add GitHub Actions workflow - #1
Merged
Merged
Conversation
- Validate project directory exists before config validation so the correct FileNotFoundError is raised for missing directories - Fix loop variable name collision in render.py (edge vs flow) - Fix llm.py last_error type annotation to Exception | None - Fix call_llm return type to CodebaseGraph | str - Use cast() for union return narrowing in __init__.py - Fix ArchitectureNote dict -> model in dry-run code path
Path.walk() was added in Python 3.13. Replace with os.walk() to maintain compatibility with the project's requires-python >=3.11.
ggrace519
added a commit
that referenced
this pull request
Sep 3, 2026
…ion #1) (#51) * feat(query): pure query layer over a generated map (innovation #1) The map's consumer is a coding agent, and a flat GRAPH.md makes it pay for the whole document to answer "who imports X?". query.py answers each such question over an already-materialized CodebaseGraph: a unified edge index (deterministic_edges ∪ import_edges, labelled ast/llm/both), fuzzy path resolution, overview, module_info, neighbors, dependents (BFS blast radius with a cap), find (ranked search across the curated sections), cycles, entry_points, staleness (stamped SHA vs HEAD), and load_map (reusing diff.load_baseline so a corrupt map is reported, not misread). No LLM, no network, no MCP dependency — this is the testable core the --serve transport wraps. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M * feat(cli): --serve exposes the map to agents over MCP (innovation #1) Adds graphlm/mcp_server.py, a stdio MCP server (SDK v2) registering eight zero-LLM tools over query.py — overview, find, module, neighbors, dependents, cycles, entry_points, staleness. GRAPH.json is re-read when its mtime changes so a regeneration mid-session is picked up; an unreadable map is raised as ToolError so the agent sees the actionable message instead of the SDK's masked "Error executing tool". The mcp package is an optional extra (graphlm[mcp]); nothing outside mcp_server.py imports it. The CLI flag short-circuits like --install-skill (single-command Typer app), defaults PROJECT_DIR to '.', honours -o for the map location, and checks the map exists BEFORE importing mcp so a missing map reports 'run graphlm .' rather than an install hint. CI now installs the extra so the in-memory-client server tests run; they importorskip without it. The --install-skill guide tells the agent to prefer the MCP tools when registered. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M * docs: document graphlm --serve (MCP) and the mcp extra (innovation #1) README section + options row + install note, CHANGELOG [Unreleased] entry, CLAUDE.md architecture paragraph (query.py/mcp_server.py split, SDK v2 ToolError rule, extra + CI), and the branch DEMO. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M * fix(query): drop question filler and match inflections in find Verified on a real map: "where is the secret redaction" ranked every quick-reference entry first because they all begin "Where is…" and the scorer counted "where"/"is" as exact-token hits. Stopwords are now stripped before scoring, and a >=4-char token matches a word it prefixes (or that prefixes it) so "redaction" reaches "_redact_secrets". Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M * fix(query): review fixes — absolute paths, distinct-file degrees, anchored refs, find shape, clamps Fresh-context review of --serve surfaced five query-layer bugs, each now with a regression test: - resolve_path could not match an absolute or repo-prefixed path (an agent's natural input) and answered "no such path" — now the longest known path that the query ends with wins; suffix candidates are capped. - in/out degree in overview and module counted edge rows, so a file that both `import x` and `from x import y` counted twice — now distinct files. - module's quick-reference lookup was an unanchored substring test (`a.py` claimed `data.py`) — now exact-file. - find returned a shape without `total` for an all-stopword query (the test had locked the bug in); stopword-only queries now search the raw tokens instead of answering nothing. - neighbors' direction is a Literal in the tool schema and ValueErrors are raised as ToolError, so the reason reaches the model instead of the SDK's masked "Error executing tool". Also: model-supplied limits clamped to MAX_LIMIT, deterministic cycle ordering on ties, and MapCache defers to load_map's classification so a broken symlink reads as unreadable rather than "run graphlm .". Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
ggrace519
added a commit
that referenced
this pull request
Sep 3, 2026
Neither the test nor typecheck job writes anything (no releases, no PR comments, no checks API calls), so scope the default GITHUB_TOKEN down to contents: read. Least-privilege, no behavior change. Fixes CodeQL actions/missing-workflow-permissions alerts #1 and #2. Claude-Session: https://claude.ai/code/session_01XhMa9VQQPB5m1Z9tP3eP4M Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add CI workflow for automated testing and type checking on every push/PR to main.
What it runs
uv, pytest with coverage reporting, Codecov uploadgraphlm/packageWhy uv
Faster dependency resolution, cached builds, and matches the project's own dev workflow.