メニュー
Diagnose GitHub Actions failures with gh by tracing failing runs, jobs, and step logs. Trigger on CI red/failing checks, flaky workflows, rerun decisions, or release pipeline breakages.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Build, launch, and drive prism's real surfaces (CLI, MCP stdio server, downstream MCP bridge, benchmarks) to verify changes end-to-end.
Diagnose Argo CD app sync/health drift with argocd (and kubectl when needed). Trigger on OutOfSync, Degraded, failed sync hooks, or unexpected desired/live diffs.
Debug Argo Workflow failures with argo CLI by tracing nodes, retries, template errors, and timing. Trigger on failed/hanging workflows, extract-step errors, or regression after upgrades.
Harvest authoritative docs for a target library/service and extract implementation-ready guidance with citations. Trigger on "how do I use/configure X" and API/reference requests.
Implement vanilla JS todo logic with localStorage persistence and deterministic state updates.
Write concise README instructions for opening and testing static HTML/CSS/JS apps locally.
| name | gh-actions-diagnostics |
| description | Diagnose GitHub Actions failures with gh by tracing failing runs, jobs, and step logs. Trigger on CI red/failing checks, flaky workflows, rerun decisions, or release pipeline breakages. |
| compatibility | Requires GitHub CLI and workflow log visibility; collector can auto-download gh into ~/.cache/prism/bin (set GH_VERSION / GH_BIN / PRISM_BIN_DIR). |
| metadata | {"prism-agents":"github-cli"} |
gh run list --limit 20gh run list --branch <branch> --limit 20gh run list --workflow "<workflow>" --limit 20gh run view <run-id>gh run view <run-id> --json jobs,conclusion,createdAt,updatedAt,headSha,event,workflowNamegh run view <run-id> --log-failedgh run view <run-id> --logtransient)deterministic/config)summary: short root-cause hypothesis with confidence.findings: failing job, failing step, first concrete error, recurrence signal.artifacts: include run/job IDs and key log snippets.confidence: high when error is explicit and reproducible; medium/low otherwise.insufficient_evidence with a callback request to parent for exact run URL/ID.