メニュー
Run the full certifier pipeline (Phase 0+1+2+3) locally against a Langfuse trace using scripts/run_certification.py, then summarize the .tmp/ output tree. Use for the certifier inner dev loop without starting the FastAPI service.
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
Advance one or all submodule pointers in the ACE monorepo to the head of their tracked branch, verify the target commit is pushed upstream, and draft a Conventional-Commit pointer-bump commit. Use when updating submodule pointers after a submodule change has merged.
Generate and run unit tests for the current diff in the correct framework per language (Go testing/table-driven, Python pytest, TS), then report pass/fail. Use before opening a PR to satisfy the "tests added/updated" checklist item.
Open a pull request following the ACE CONTRIBUTING conventions — Conventional-Commit title, the standard PR description template, and auto-detected affected submodules / pointer bumps. Use when the user is ready to raise a PR.
Run the documented certifier end-to-end smoke test against the running FastAPI service (bucketing-extraction → poll → aggregation-certification → poll). Use to verify the certifier API path works before opening a PR.
Build and push all ACE component Docker images to Docker Hub via scripts/build-and-push.sh, confirm the five images pushed, and report the new certifier:latest digest. Use to publish a new release of the images.
Bootstrap a local ACE dev environment — copy and validate .env / build-paths.env, then bring up MongoDB, Langfuse, LiteLLM, and the Certifier via start-local-services.sh. Use when onboarding, after a fresh clone, or when local services are not running.
| name | run-certification |
| description | Run the full certifier pipeline (Phase 0+1+2+3) locally against a Langfuse trace using scripts/run_certification.py, then summarize the .tmp/ output tree. Use for the certifier inner dev loop without starting the FastAPI service. |
Exercise the certifier pipeline end-to-end against one trace, off the API path.
.env filled in with LANGFUSE_* and AZURE_OPENAI_* (the script reads them directly).reportlab (pip install --user reportlab)../scripts/run_certification.py --trace-id <LANGFUSE_TRACE_UUID>
./scripts/run_certification.py \
--agent-id <UUID> --experiment-id <UUID> --run-id <UUID> --agent-name vaya
--skip-cert (stop after Phase 0+1), --no-pdf (skip PDF render), --runs-per-fault N (Phase 2 significance), --batch-size N (LLM classification batch), --workspace <dir> (default .tmp/), --debug (retain Phase 3 intermediates)..tmp/{agent_id}/{experiment_id}/. Report the key artifacts:
fault-bucketing/{run_id}/metrics/*_metrics.jsonaggregation/aggregation.jsoncert-builder/certification.json (final certificate)cert-builder/certification.pdf (rendered report).tmp/ is gitignored — safe to leave artifacts there.certification.json, use scripts/render_certification_pdf.py.pipeline-smoke-test skill.