From cornerman
Aggregate cornerman telemetry, run manifests, and incidents into a markdown retro report. Read-only — no mutations. Surfaces where PRs stall, which contracts fail most, stuck issues, ci-fixer effectiveness, token outliers, recent watchdog incidents, and verification drift. Use when asked to "show retro", "loop summary", "where's the loop stalling", or weekly to spot patterns.
How this skill is triggered — by the user, by Claude, or both
Slash command
/cornerman:cornerman-retroThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Read-only telemetry aggregation. Produces one markdown report from the local `.claude/loop-events.jsonl` + `.claude/runs/*/manifest.json` + `.claude/incidents/*/`. Doesn't mutate anything.
Read-only telemetry aggregation. Produces one markdown report from the local .claude/loop-events.jsonl + .claude/runs/*/manifest.json + .claude/incidents/*/. Doesn't mutate anything.
| Flag | Default | What it does |
|---|---|---|
--since | 7d | 7d, 24h, 30d, all |
--issue <N> | none | Filter to a single issue |
--agent <name> | none | Filter to a single agent (implementer, pr-reviewer, …) |
The skill invokes:
python3 .claude/plugins/cornerman/scripts/retro/aggregate.py \
--since "${SINCE:-7d}" \
${ISSUE:+--issue "$ISSUE"} \
${AGENT:+--agent "$AGENT"}
Output is markdown to stdout. Sections:
state_transition event (heuristic; tunable in the aggregator)When .claude/loop-events.jsonl doesn't exist (or filters exclude everything), the report says so explicitly — does NOT print empty sections. Same shape as research-platform-scan surfacing missing platform_capabilities (#2): visible gap beats silent nothing.
/cornerman where-am-i (#14) for per-issue / per-branch resumability/cornerman resume (#22) for full session restoration with what-changed-since/cornerman retro
/cornerman retro --since 24h
/cornerman retro --issue 211
/cornerman retro --agent implementer --since 30d
/cornerman retro --since all # since beginning
scripts/retro/aggregate.py ships with the plugin. Pure Python (no jq dependency); reads YAML/JSON only. ~5KB. Stand-alone — runnable outside the skill for ad-hoc reports.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub boutlabs/claude-plugins --plugin cornerman