Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

/retro — CP-7 close the V — measured AI agent skill | DecimalAI

huytieu/retro

NewSkillScoreMaintained

CP-7 retrospective: audit checkpoints, evidence quality, action items, and harvest candidates. Closes the V-model cycle and feeds the next run. Use via /retro after ship, escalate, or significant session.

agent-toolingby huytieu

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

Project Management pack

decimalai skills pull huytieu-retro

Get this skill — curl · raw file

curl -fsSL https://app.decimal.ai/s/huytieu-retro/SKILL.md

Raw file: /s/huytieu-retro/SKILL.md ↗ · machine-readable summary at /s/huytieu-retro.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

No production data for this skill yet — usage numbers appear here once agents start running it.

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 3h ago

—-10% tokens0% turns64% of whole cases pass with skill (14/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-08	✗→✓	▲ Improved	-9%	0%
case-14	✗→✓	▲ Improved	27%	0%
case-15	✗→✓	▲ Improved	2%	0%
case-16	✗→✓	▲ Improved	-23%	0%
case-17	✗→✓	▲ Improved	-22%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

huytieu / cog-second-brain

.agents/skills/retro/SKILL.md

View on GitHub ↗

License MITImported 8/10/2026Version v2

Skill contents

SKILL.md

/retro — CP-7 close the V

Every non-tiny shipped or escalated run gets a retro. Retro is advisory but strongly expected: it is how the harness improves.

When to run

After CP-6 ship (approved publish, merged PR, vault deliverable done)
After escalate (capture why verifier failed)
End of an /ultragoal phase (one retro per phase)
Manual: /retro <run-dir or spec path>

Phase 1 — Gather evidence bundle

Read:

Spec or card description + acceptance criteria
04-projects/harness/runs/<id>/evidence/ (ledger, checkpoints.tsv, CP- files)
.claude/logs/loop-ledger.tsv (last rows for this run)
04-projects/<goal>/STATUS.md if an ultragoal phase

Phase 2 — Checkpoint audit

For each CP the lane required, mark Expected vs Actual vs Gap:

| CP | tiny | normal | full | |---|---|---|---| | CP-1 spec | skip | required | required | | CP-3v component | skip | required | required | | CP-4 integration | skip | skip | required if multi-task | | CP-5 acceptance | if mutate | required | required | | CP-6 ship | if external | if external | required |

Phase 3 — Evidence quality pass

Ask:

Any AC-n shipped without a PASS evidence row? (traceability violation)
Any PASS row where observation is tool-return not artifact? (confident-but-unchecked)
Any criterion that wasn't falsifiable? (fix in next spec)

Phase 4 — Write retro

Copy 04-projects/harness/templates/retro.md to:

04-projects/harness/retro/YYYY-MM-DD-<slug>.md

Fill all sections. Action items get IDs (AI-01…).

Phase 5 — Feed forward

Harvest: append action items + lessons to today's harvest staging
Backlog: add AI-n rows to 04-projects/harness/BACKLOG.md if harness work
STATUS.md: if an ultragoal phase, advance the phase state + log open AC-n
Spec checkpoint log: update spec ## Checkpoint log CP-7 row
Record: bash .claude/lib/checkpoint.sh record <run-dir> CP-7 PASS "retro: <path>"

Output to the user

TL;DR: outcome, top lesson, top action item, retro path.

Want to see if huytieu/retro would activate on your phrasing? Try it before you install — no LLM call, no signup.

Other measured skills in the registry, with their headline benchmark lift.