Skip to content

feat: add /interview command for focused questioning#18

Merged
leeovery merged 1 commit intomainfrom
claude/improve-feature-research-questions-1y5ET
Dec 29, 2025
Merged

feat: add /interview command for focused questioning#18
leeovery merged 1 commit intomainfrom
claude/improve-feature-research-questions-1y5ET

Conversation

@leeovery
Copy link
Copy Markdown
Owner

Adds a new slash command that shifts into focused questioning mode
during research or discussion phases. Uses AskUserQuestion tool to
probe ideas with non-obvious questions, challenge assumptions, and
surface concerns early.

Adds a new slash command that shifts into focused questioning mode
during research or discussion phases. Uses AskUserQuestion tool to
probe ideas with non-obvious questions, challenge assumptions, and
surface concerns early.
@leeovery leeovery merged commit a55af27 into main Dec 29, 2025
@leeovery leeovery deleted the claude/improve-feature-research-questions-1y5ET branch December 29, 2025 12:00
leeovery added a commit that referenced this pull request Apr 12, 2026
Evaluated Claude Code /loop and channels features for enhancing
background agent surfacing. Conclusion: loop as periodic safety net
has marginal value — needs real-world observation before committing.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
leeovery added a commit that referenced this pull request Apr 16, 2026
No automatic catch-up for failed knowledge remove calls — unlike
indexing which has a pending queue. Stale chunks persist until
manual retry. Logged for future mitigation.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
leeovery added a commit that referenced this pull request Apr 20, 2026
knowledge remove previously had no retry — a locked store or transient
I/O failure left stale chunks in the index forever and told the user to
retry manually. Indexing already has a pending queue; removals now
mirror it.

Changes:
- metadata.pending_removals tracks failed removals with workUnit/phase/
  topic, attempts counter (cap 10, then evict with stderr warning).
- cmdRemove drains the queue first, attempts the new removal, queues on
  failure.
- cmdCompact drains the queue on entry.
- status surfaces pending removals alongside pending indexes.
- Skill callsites (cancellation, supersession, promotion) updated to
  describe auto-retry instead of 'run knowledge remove manually later'.

Closes deferred-issues #18.
leeovery added a commit that referenced this pull request Apr 21, 2026
The pending-removal queue (deferred-issue #18 fix, commit b5c9430)
shipped broken. writeMetadata whitelists a fixed 5-field schema and
pending_removals was never added — every write silently stripped the
queue. addPendingRemoval wrote the entry, next readMetadata+writeMetadata
cycle dropped it, processPendingRemovals found nothing to drain.

Automatic retry on failed remove (cancellation / supersession /
promotion paths) therefore never happened. Stale chunks from failed
removes persisted forever, contradicting the user-facing promise in
manage-work-unit.md, spec-completion.md, promote-to-cross-cutting.md.

Changes:
- Add pending_removals to writeMetadata's whitelist (store.js:443-444).
- Update the whitelist comment with a loud warning that missing a field
  silently breaks every downstream feature using it.
- METADATA_FIELDS constant updated for consistency.
- Test 81 rewritten: Part A is now a true regression guard — seeds a
  pending_removals entry, runs an index (which triggers writeMetadata),
  asserts the entry survives. This assertion fails on the pre-fix code,
  confirmed by temporarily reverting the whitelist. Part B retains the
  drain-the-queue check.

All existing tests still pass (142/142 CLI, integration, smoke).
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants