Pipeline retrospective and pattern extraction for continuous learning. Use after every pipeline completion to capture lessons.
Pipeline retrospective and pattern extraction for continuous improvement.
Examine the completed pipeline run end to end:
Look for recurring themes across this and previous runs:
Create or update pattern files with structured observations:
Keep the project knowledge base current:
Read the template before producing the retrospective:
Read ${CLAUDE_PLUGIN_ROOT}/templates/RETRO.md
Patterns earn confidence through repeated observation:
| Event | Score Change |
|---|---|
| First observation | 0.3 |
| Each additional observation | +0.2 |
| Maximum confidence | 0.9 |
| User rejects the pattern | -0.1 |
A pattern at 0.3 is a tentative observation. A pattern at 0.7+ is well-established and should be actively applied in future runs.
Pattern and context data is stored per-project:
${COMPANY_OF_ONE_PLUGIN_DATA}/projects/{project-key}/
pipeline.json # Flow control (stage, status, timestamps)
context.md # Project knowledge (tech stack, conventions)
patterns/ # Extracted patterns with confidence scores
index.json # Pattern index (session-start reads only this)
briefs/ # Agent handoff (current.json + history/)