fix(hooks): Parse Claude Code's nested tool_input payload

[JordanCoin]

Closes #110.

The bug

parseHookFilePathsErr handled a top-level file_path and Codex's tool_input.command, but not tool_input.file_path — the shape Claude Code actually sends for Edit and Write.

The regex fallback that would have caught it only runs when json.Unmarshal fails. A well-formed payload in the unhandled shape parses cleanly, matches nothing, and returns no paths — so the hook returned early and did nothing, silently.

Same file, same hook, before this change:

$ printf '{"session_id":"s1","file_path":"<abs>/cmd/doctor.go"}' | codemap hook post-edit
   Imports 1 hub(s): internal/buildinfo/version.go
  recorded: 1

$ printf '{"session_id":"s1","tool_name":"Edit","tool_input":{"file_path":"<abs>/cmd/doctor.go"}}' | codemap hook post-edit
  recorded: 0        # no output at all

After:

$ printf '{"session_id":"s1","tool_name":"Edit","tool_input":{"file_path":"<abs>/cmd/doctor.go"}}' | codemap hook post-edit
   Imports 1 hub(s): internal/buildinfo/version.go
  recorded: 1

Impact

Two features were inert on Claude Code:

• Agent-edit provenance — .codemap/agent_edits.jsonl was never created, so the working-set summary always fell back to disk churn. That conflation is precisely what the feature was built to remove. Across a four-day session in this repo with a dozen-plus Edit/Write calls and four merged PRs, the file never appeared.
• Post-edit blast radius — the "you just edited a file N others import" report never fired, which is the behavior CLAUDE.md tells agents to rely on.

Also now accepts tool_input.notebook_path for NotebookEdit.

Tests

Six payload shapes, covered end to end rather than only at the parser — the parser was arguably correct, it just never saw the shape it needed, so a parser-only test would have passed while the feature stayed broken:

Shape	Expect
legacy top-level file_path	resolves
tool_input.file_path (Edit)	resolves
tool_input.file_path (Write)	resolves
tool_input.notebook_path (NotebookEdit)	resolves
tool_input.command (Codex apply_patch)	resolves
tool_input.command (Bash, no path)	no match

Plus TestHookPostEditRecordsAgentEditForNestedPayload, which drives hookPostEdit and asserts the record lands.

Docs

Updated now that the behavior actually works, rather than documenting the broken state:

• docs/HOOKS.md — describes working-set provenance and the payload keys it depends on; notes doctor's project→user scope fallback from fix(doctor): Report effective agent config, not file layout #100; the post-edit row now mentions provenance.
• The manual hook JSON deliberately keeps the bare PATH form — that's correct for hand-managed settings and doctor has accepted it since fix(doctor): Report effective agent config, not file layout #100.
• Drive-by: CONTRIBUTING.md offered Elixir and Scala as languages "to add" (both shipped); plugins/codemap/README.md said "skill" singular but ships two.

Type of change

• Bug fix
• New feature
• New language support
• Documentation
• Other

Verification

go test ./..., go vet, staticcheck, gofmt clean; go vet clean cross-compiled for windows/amd64 and linux/amd64. Behavior confirmed against a built binary with the exact payload Claude Code emits.

[Copilot]

Pull request overview

Fixes the PostToolUse hook path extraction so Claude Code payloads for Edit/Write/NotebookEdit are correctly recognized, restoring agent-edit provenance recording and the post-edit blast-radius reporting that depends on it.

Changes:

• Extend hook payload parsing to accept tool_input.file_path and tool_input.notebook_path in addition to the existing legacy and Codex shapes.
• Add end-to-end and parser-level tests covering multiple host payload shapes, including a regression test that drives hookPostEdit.
• Update hook and contribution documentation to reflect the now-working provenance behavior and current project state.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
plugins/codemap/README.md	Fix wording to reflect multiple shipped skills.
docs/HOOKS.md	Document provenance-based working set behavior and clarify doctor scope fallback behavior.
CONTRIBUTING.md	Update example “languages to add” list to avoid listing already-supported languages.
cmd/hooks.go	Parse nested tool_input.file_path / tool_input.notebook_path for Claude Code hook payloads.
cmd/hooks_payload_test.go	Add regression tests for supported payload shapes and nested provenance recording.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.