journal index doesn't cover voice notes — the 'run journal index to index later' recovery hint is a dead end

[claude]

🤖 Agent-Generated PR

Closes #118
Agent Used: Claude Code
Complexity: medium

---

Summary

journal index never scanned the voice-note landing zone (log.landing.dir, default logs/), so when land-time indexing failed (e.g. Ollama down) the printed recovery hint — "run journal index to index later" — was a dead end: the note stayed on disk, unindexed, forever. journal index now scans that landing zone and indexes voice notes the same way the land-time path does.

Implementation Notes

• Added index.WalkVoice(root, logRel, since) in internal/index/walk.go, sharing the existing directory-walk logic with WalkTranscripts via a new private walkLandingDir helper (dedup, not new abstraction — the two walks were already structurally identical). Voice notes only ever land as .md (see internal/log.Filename), so WalkVoice matches .md only.
• runIndex (cmd/index.go) now walks the voice landing zone after the notes/transcripts passes and indexes each file via internal/log.IndexVoice (the same helper journal log calls at land time), reusing the indexer's stable content-hash diffing — unchanged voice notes are a no-op on re-index, and --rebuild re-embeds them like everything else.
• noteExcludes now also excludes the voice landing zone from the plain-note walk, mirroring the existing transcripts exclusion. Without this, the voice-note .md file would also get picked up by the regular note walk and re-chunked (by heading) as source=note — a second, differently-chunked copy of the same content living alongside the correct source=voice chunks. Unfiltered journal search already matches every source (Sources: nil = any), so voice notes remain fully searchable without that duplicate.
• Scope: this only changes the non-watch journal index command, per the issue and acceptance criteria. journal index --watch doesn't drive voice notes through IndexVoice either before or after this change; a landed-while-watching voice note is picked up by the next plain journal index run. Extending live --watch coverage to voice notes is a separate concern and out of scope here.
• No changes to chunking or the land-time indexing path (out of scope per the issue).

Changes

• internal/index/walk.go — new WalkVoice (voice landing zone walk), refactored shared logic into walkLandingDir.
• cmd/index.go — runIndex indexes voice notes via IndexVoice; noteExcludes excludes the voice landing zone from the plain-note walk.
• cmd/index_test.go — new tests (below).
• CHANGELOG.md — entry under the in-progress 3.0.2 section.

Testing

Automated:

• Unit tests added/updated
• All existing tests pass (go test ./..., go vet ./..., gofmt -l . clean)

New tests in cmd/index_test.go:

• TestRunIndexRecoversUnindexedVoiceNote — reproduces the issue: force land-time index failure, then run journal index and confirm the note becomes searchable via --source voice; also asserts the plain-note walk scanned 0 files (voice note not double-indexed as source=note).
• TestRunIndexVoiceNoopReindex — re-running journal index on an already-indexed, unchanged voice note makes zero additional embed calls.
• TestRunIndexRebuildReembedsVoiceNotes — journal index --rebuild re-embeds voice notes and they remain searchable via --source voice.

Manual testing performed:

• Ran the full suite locally (go test ./...), go vet ./..., and gofmt -l . — all clean.

---

Review Checklist

Correctness

• Code actually solves the issue
• Acceptance criteria are met
• Edge cases handled

Code Quality

• Follows existing codebase patterns
• Follows project coding standards
• Naming conventions consistent
• No unnecessary complexity
• Comments where needed (but code is self-documenting)

Security

• Input validation present (n/a — no new external input)
• No exposed secrets or credentials
• Data properly sanitized/escaped (n/a)
• Authentication/authorization checks where needed (n/a)

Performance

• No obvious N+1 queries
• Reasonable algorithmic complexity
• Caching used where appropriate (n/a)
• No blocking operations in request lifecycle (n/a — CLI command)

Tests

• Tests actually test behavior (not just coverage)
• Edge cases covered
• Failure modes tested

---

Reviewer Notes

The decision to exclude the voice landing zone from the plain-note walk (so voice notes aren't double-indexed under both source=note and source=voice) is a small scope expansion beyond the letter of the issue, but is necessary for correctness — without it, journal index would silently create a second, wrong-shaped copy of every voice note's content. Flagging this explicitly for review.

[claude]

Retro: no durable lesson (merged clean / one-off).