| name | small-feature |
| description | Lightweight workflow for small refinements — tweaks, config changes, prompt tuning, single-module changes |
| wiki_tiers | ["essential","patterns"] |
Small Feature / Refinement
For changes touching few files, refining existing systems, no deep research needed.
Phase A — Quick Research
Output: Findings in chat (no doc unless complexity surprises).
- Read relevant
README.md (per-folder docs first).
- Parallel scan — 2
Explore agents:
- Agent 1: Read primary files — current behavior, data flow
- Agent 2: Read callers, subscribers, consumers — ripple risks
Consult external API/library docs if relevant — don't guess.
- Summarize: what exists, what changes, risks.
- Surface ambiguities. If clear, propose plan.
Phase B — Mini Plan
Output: Plan in chat (no doc). Skip if obvious.
- List: files to modify/create, types, acceptance criteria. Mark independent files.
- Flag risks: dependency direction violations, lifecycle mismatches, event chain hops.
- Get go-ahead.
Phase C — Implement
Read STANDARDS.md first.
- Independent files in parallel agents. Sequential only for deps.
{{BUILD_CHECK_CMD}} → summarize.
- Scope beyond ~3 files → pause, discuss.
- CLAUDE.md Verify Checklist — every box. Plus:
- Grep callers of changed exports — behavioral correctness
- Trace error path of every new async call
{{BUILD_CHECK_CMD}} final verify.
- Update memories if architecturally notable.