| name | finish |
| description | Quality gate before marking any task complete. Runs cs-fixer + full tests + coverage, then updates shared agent memory with lessons learned. |
Finish Protocol
Run this skill before declaring any task complete.
Steps
1. Run the quality gate
./docker.sh test-all-coverage
Note: this command can take 5–15 minutes on first run or after container rebuild.
This runs (in order):
./docker.sh cs-fixer — fix all code style violations- Full unit test suite with coverage report
If this fails: Stop. Report what failed. Do NOT update memory. The task is not done — fix the failure first.
If this passes: Continue to step 2.
2. Review shared memory
Read .claude/memory/MEMORY.md to see what memory files exist, then read any that are relevant to the work you just completed.
3. Update memory
Ask yourself: "Did I learn anything non-obvious during this task?"
Examples of what belongs in memory:
- A class/method that behaves unexpectedly
- A pattern that works well (or poorly) in this codebase
- A pitfall that caused a bug or wasted time
- An architectural constraint that wasn't obvious from reading the code
If yes: Find the relevant memory file and append the lesson. If no file fits, create a new one with frontmatter and add it to MEMORY.md.
If no: Skip this step.
Memory file frontmatter format:
---
name: <name>
description: <one-line summary used to decide relevance>
type: reference | feedback | project
---
4. Report to the user
Summarise:
- Tests passed (or what failed)
- Coverage report location:
coverage/html/index.html
- What (if anything) was written to memory
