菜单
Read the committed Actions analytics (_data/actions_usage.yml), explain the worst workflows (cost, effectiveness, waste), and drive the next action — dispatch the actions-review issue-filer or fix a workflow directly. Use when asked "where are Actions minutes going" or after the daily actions-usage refresh.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Deep-analyze a GitHub issue into a client-engagement estimate — classify scope, refine the plan (approach/deliverables/acceptance), and record an agent-refined estimate in _data/engagements.yml. Use when asked to "estimate", "quote", "scope", or "price" an issue or a repo's backlog, or to review/refine a draft the estimator produced.
Read the Monitor board health data, explain why each repo needs attention, and propose the next concrete action per repo. Use to turn monitoring signals into action, or as the input step of the self-evolution loop.
Run the dash drift gate and explain every failure in plain language with the exact fix. Use when CI drift-check fails or before opening a PR.
Run a focused improvement pass on a single project (code quality, docs, tests, error handling) using the matching .github/agents persona. Use when a repo needs maintenance attention or as the action step of the evolution loop.
Scaffold a new project and register it in the dash. Use when starting a new repo/app that should appear in the portfolio.
Adopt (or remove) a project directory that exists on disk under projects/ but is in neither .gitmodules nor the registry. Use when the drift gate flags a stray/unregistered project dir.
| name | actions-triage |
| description | Read the committed Actions analytics (_data/actions_usage.yml), explain the worst workflows (cost, effectiveness, waste), and drive the next action — dispatch the actions-review issue-filer or fix a workflow directly. Use when asked "where are Actions minutes going" or after the daily actions-usage refresh. |
The analytics loop: tools/dash-gen actions (actions_analytics.py, PyGithub) writes the committed _data/actions_usage.yml, refreshed daily by actions-usage.yml; actions-review.yml is the downstream AI issue-filer. This skill is the local triage between the two.
_data/actions_usage.yml. Check generated_at / window_days; if stale or missing, regenerate with tools/dash-gen actions (needs a GitHub token — GH_TOKEN/GITHUB_TOKEN or gh auth login — plus PyGithub).workflows list is pre-sorted by priority (waste minutes + total minutes discounted by effectiveness). Per entry: repo, workflow, type, runs, total_min, avg_min, p95_min, waste_min, effectiveness_pct (share of minutes ending in success), success_rate_pct, sched_pct, events, and flags (high-cost-low-value, failing, flaky, slow, cancel-heavy, cron-heavy). Rollups live in totals, by_type, by_repo, by_day; inactive lists workflows that never ran in the window (prune candidates).actions-review-labeled issues to avoid duplicates: gh issue list -R bamr87/bamr87 --label actions-review --state open.concurrency group with cancel-in-progress (cancel-heavy), timeout-minutes (slow), paths/paths-ignore filters, a looser cron cadence (cron-heavy), or disabling/pruning an inactive workflow.gh workflow run actions-review.yml (deterministic candidate selection + dedupe, then an Opus reviewer files one optimization issue per candidate; dry_run / max_issues inputs)._data/actions_usage.yml is GENERATED (its header says so) — never hand-edit it; fix the generator or the workflows it measures.