Continuous improvement feedback loop that captures observations during work (/note), analyzes root causes with 5 whys (/review), and implements durable improvements via the claude-code-optimizer (/retro). Triggers on retrospective, retro, improvement, feedback loop, root cause, 5 whys, observation, friction.
From dot-claudenpx claudepluginhub selrahcd/claude-marketplace --plugin dot-claudeThis skill uses the workspace's default tool permissions.
retro-process.mdsolution-hierarchy.mdDesigns and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Every friction moment is an improvement opportunity. This skill powers a three-phase feedback loop:
/note) — Log observations during work. Quick, low-friction./review) — Scan the conversation for unlogged issues with 5 whys root cause analysis./retro) — Process all pending observations, delegate to claude-code-optimizer to find the best solution, then implement immediately.The output isn't a document. It's working improvements to skills, commands, CLAUDE.md, AGENTS.md, hooks, linter rules, etc.
All files in ~/.claude/retro/:
observations.md — Pending and processed observationslog.md — Append-only audit trail of improvements madeEach observation in observations.md follows this template:
## [YYYY-MM-DD HH:MM] Title
- **Status:** pending | addressed | skipped
- **Project:** project-name
- **Context:** What was happening when this was observed
- **Details:** The observation itself
- **Root cause:** (filled by /review, empty for /note captures)
- **Resolution:** (filled by /retro when addressed)
Each entry in log.md follows this template:
## [YYYY-MM-DD HH:MM] Improvement title
- **Source:** observation title or conversation finding
- **Root cause:** Why this happened
- **Solution:** What was implemented
- **Type:** skill | command | claude-md | hook | linter | test | type-system
- **Files changed:** list of modified files
During normal work (not just during /note), you may suggest capturing an observation when you spot clear failures:
Ask: "Want me to note this for review?" — only log if user agrees. Never log silently. Never suggest for minor issues. One prompt, then move on.