Runs after every track completion to extract learnings. Analyzes what worked, what failed, and what patterns emerged. Updates conductor/knowledge/patterns.md with new solutions and errors.json with new error patterns. Proposes skill updates if workflow improvements are identified. Triggered automatically by orchestrator after COMPLETE step.
SkillScore86Maintained
Automatic
◆Knowledge Management pack
Last verified 8/23/2026
What it changes
Passed 16 of 18 test cases with the skill, and 11 of 18 without it.
Without skill59%
With skill86%
Checks it passed that the model alone missed
The response names skill_improvements as the JSON key
The response specifies conductor/knowledge/patterns.md as the destination path
The response identifies spec.md as the input used to evaluate requirement alignment
and 4 more in the full report
6 fixed13 passed either way3 still failing0 broke
Pass-rate lift
+27 pp
Tokens
+115%
Agent turns
0%
Response time
-35%
Model gemini-3.6-flashMeasured 8/23/2026v119 of 22 cases passed with skill · 18 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
What key name in the retrospective agent's final output JSON structure holds the list of proposed workflow and evaluator enhancements? Is it `action_items`, `workflow_fixes`, or `skill_improvements`?
What was checked, without and with the skill
Checked
Without skill
With skill
The response names skill_improvements as the JSON key
Missed
Passed
case-11 · gemini-3.6-flash · 8/23/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md288 linessingle file
Retrospective Agent — Post-Track Learning
Extracts learnings from completed tracks and updates the knowledge base, making every future track smarter.
When This Runs
Automatically — The orchestrator calls this agent AFTER a track reaches COMPLETE status.
Inputs
Track's plan.md — All tasks, including fix cycles
Track's metadata.json — Fix cycle count, lead consultations, blockers
Track's commits — What was actually built
Track's spec.md — Original requirements (to measure alignment)
Workflow
1. Analyze Track Execution
Review the track's execution data:
## Track Analysis: [track-id]
### Execution Summary
- **Total Tasks**: 12
- **Fix Cycles**: 1
- **Lead Consultations**: 3
- **Duration**: Plan to Complete
### What Worked Well
- [Extracted from smooth tasks with no fix cycles]
### What Caused Problems
- [Extracted from fix cycles and blocked tasks]
### Patterns Discovered
- [New solutions that could be reused]
### Errors Encountered