Feature: index GitHub Actions workflows (.github/ is hard-excluded; no GHA-aware parser)

[ilyabrykau-orca]

Summary

.github/ is hard-excluded from indexing, so GitHub Actions workflows, composite actions, and any infra/config under .github/ are never indexed. There is also no GHA-aware parsing (no jobs/steps/uses graph). CI/CD is a meaningful part of most repos' architecture and is currently invisible to the graph.

Version

codebase-memory-mcp 0.8.1 (macOS arm64)

Evidence

Reindexed 4 repos with mode: full. Every project reports .github (or .github/scripts) in the excluded dirs list, e.g.:

project: ...-orca         excluded.dirs: [..., ".github", "devops", "system_tests", ...]
project: ...-orca-runtime-sensor   excluded.dirs: [..., ".github/scripts", ...]

Post-index graph query confirms zero nodes:

MATCH (n) WHERE n.file CONTAINS '.github' RETURN n.file, labels(n) LIMIT 20
-- total: 0  (on every project)

The exclusion appears hardcoded in the binary — not overridable:

• config list / config get exposes no exclude/ignore/include keys (only auto_index, auto_index_limit).
• No CBM_* env var controls directory exclusions (checked CBM_CACHE_DIR, CBM_WORKERS, CBM_LSP_*, CBM_SEMANTIC_*, etc. — none cover excludes).

Note: generic YAML is supported when reachable — binary strings reference .yaml/.yml, Chart.yaml, kustomization.yaml, compose.yaml, Dockerfile, .hcl/.terraform. So the blocker for workflows is purely the .github directory exclusion + lack of GHA schema awareness, not YAML support itself.

Requested

1. Stop hard-excluding .github/ (at minimum .github/workflows/ and .github/actions/), OR make the deny list overridable via config/env/.codebasememoryignore-style allowlist.
2. GHA-aware parsing (nice-to-have): model workflows/jobs/steps as nodes and uses: references as edges (action → action, workflow → reusable-workflow) so call/impact tracing covers CI/CD.

Why it matters

• Infra-as-code and deploy logic increasingly lives in .github/ (reusable workflows, composite actions, deploy steps calling terraform).
• Impact analysis ("what breaks if I change this action/secret/env") is impossible while CI is unindexed.
• A general escape hatch for the hardcoded deny list would also unblock other intentionally-excluded dirs (devops/, system_tests/, etc.).

[ilyabrykau-orca]

Closing — premise was wrong, my error.

I claimed .github is hard-excluded by the binary. It is not. Re-checked:

• strings on the binary shows no .github deny literal (only the api.github.com update-check URL).
• 3 of 4 of my repos index .github/workflows/* correctly (104 / 339 / 36 nodes). Example node: .github/workflows/terraform.yaml → jobs.
• The one repo with 0 .github nodes excludes it via its own .cbmignore (.github/), not the binary.

My "0 nodes" reading came from querying the wrong node property (n.file instead of n.file_path) — a false negative on my side. CBM respects .cbmignore/.gitignore as documented and does index GitHub Actions workflows. Sorry for the noise.