agent-ops-retrolisted

Retrospective on how the agents themselves are being operated, from local Claude Code transcripts plus the reports other jobs already produce. Answers what is going well, what is going wrong, what the human keeps having to repeat, and where delivery is leaking, then carries every unactioned finding forward with its age. USE THIS AUTOMATICALLY (no need for the user to name it) whenever the user asks to check all conversations, review agent usage, reduce the manual administration of agents, find repeating corrections or issues across sessions, audit skills and the knowledge ledger together, or asks what gaps exist in delivery over a recent window. This is the operator axis: for repository history and shipped work use retro-analysis, for a single finished change use wrap-up.
oleg-koval/agent-skills · ★ 3 · AI & Automation · score 75

Install: claude install-skill oleg-koval/agent-skills

# Agent Ops Retro A retrospective on the operating layer, not the code. It asks whether the agents are being run well: what the human had to repeat, what the guardrails caught, what they cost, and which findings have been reported before and never actioned. ## Scope boundary Three retro skills exist and they do not overlap. | Skill | Looks at | Answers | |---|---|---| | `retro-analysis` | Repository history, delivery evidence, code quality | What shipped and how the code is trending | | `wrap-up` | One finished change | Did this specific piece of work meet its goal | | `agent-ops-retro` | Transcripts, hooks, skills, ledger, delegation logs | How the agents are being operated | If the user asks about shipped work, hand off to `retro-analysis` and say so. ## Invocation Accept a window: `14d` (default), `7d`, `30d`, or an explicit ISO date range. State the absolute range at the top of the output. Convert relative dates to absolute. ## Step 0. Read, do not recompute Several jobs already report on this state. Adding a fifth independent opinion is the failure mode this skill exists to fix, so read their outputs first and only compute what none of them cover. | Source | Path | Gives you | |---|---|---| | Skill audit | `~/.claude/skill-audit/latest.json` and the dated `.txt` beside it | Installed vs never-invoked skills | | Delegation log | `~/.claude/delegation-metrics/delegations.jsonl` | Model, tier,