scaffold/.claude/skills/retro/SKILL.md
Retrospective entry format for capturing lessons learned at task close-out
npx skillsauth add gstredny/lauren-loop retroInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Append a retrospective entry to docs/tasks/RETRO.md when closing out a task. Entries capture what was learned — not what was done.
At task close-out only, after the user says "close it out" and all Done Criteria are confirmed. Write the entry before moving the task file to docs/tasks/closed/.
Append to docs/tasks/RETRO.md. Never edit or delete previous entries — the log is append-only.
### [YYYY-MM-DD] Task: [task name]
- **What worked:** [approaches worth repeating]
- **What broke:** [failures to avoid next time]
- **Workflow friction:** [process issues, not code issues — or "None"]
- **Pattern:** [generalizable lesson for future tasks]
The Pattern field is the most important part. It should be a generalizable lesson that applies beyond this specific task.
Good patterns:
Bad patterns:
A good test: could someone working on a completely different task benefit from reading this pattern? If yes, it's generalizable. If no, rewrite it.
A pre-tool-use hook blocks moving task files to docs/tasks/closed/ unless a matching retro entry exists in docs/tasks/RETRO.md. The entry must contain the task name in its heading.
development
Follow the end-to-end delivery workflow for planning, task-file management, execution, code review, verification, and closeout without relying on Claude-specific hooks or command conventions.
testing
Plan implementation work before coding by defining public interfaces, real collaborator boundaries, and vertical RED or VERIFY slices.
development
Manage task files in docs/tasks/open and docs/tasks/closed, including task discovery, resume/create rules, append-only Attempts logging, Left Off At updates, Code Review field handling, and needs verification closeout.
data-ai
Review an implementation plan before execution to catch missed success criteria, scope creep, constraint violations, unverified assumptions, and missed reuse opportunities.