菜单
Evaluate-phase skill for pi-forge. Use during the Evaluate phase to make a holistic, freeform judgment of the work against the brief, write artifacts/scorecard.md with an approve|iterate recommendation plus justification and risks, and record the verdict via forge_metric to inform the G2 gate.
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。
基于 SOC 职业分类
Test-phase skill for pi-forge. Use during the Test phase to run the project's test suites (unit, integration, contract, E2E for the general path), capture results and coverage, write artifacts/test-report.md, and record metrics via forge_metric.
Safe bridge to the cxas CLI for pi-forge CXAS projects. Use whenever a CX Agent Studio project needs to lint an app directory, run evaluations, manage apps/deployments, or migrate from Dialogflow CX. Always go through the forge_cxas tool (never raw shell); it enforces an allow-list and returns structured results.
First Build-phase skill for pi-forge. Use at the start of the Build phase to turn the approved brief into an executable plan. Reads artifacts/brief.json and writes artifacts/plan.md with the approach, module/feature breakdown (WBS), build sequence, and stack rationale.
Drives a project through the pi-forge delivery lifecycle (intake → clarify → build → test → evaluate → improve → deliver). Use when starting or continuing a pi-forge project, or whenever asked to "run the project", "continue the project", or after /forge-new. Reads state via forge_status and advances phases via forge_advance, opening gates with forge_gate.
Improve-phase skill for pi-forge. Use during the Improve phase after an iterate verdict to root-cause the scorecard findings, write a prioritized artifacts/backlog.md, and hand back so the orchestrator re-enters the Build phase (bounded by maxIterations).
Clarify-phase skill for pi-forge. Use during the Clarify phase to turn a raw request into a high-quality Project Brief. Scans for ambiguity across functional, non-functional, integration, data, user, edge-case, metric, and scope dimensions, asks batched categorized questions, and writes artifacts/brief.md + brief.json.
| name | evaluator |
| description | Evaluate-phase skill for pi-forge. Use during the Evaluate phase to make a holistic, freeform judgment of the work against the brief, write artifacts/scorecard.md with an approve|iterate recommendation plus justification and risks, and record the verdict via forge_metric to inform the G2 gate. |
Make a holistic judgment of the work. The output recommendation informs the human's G2 decision. Drive transitions only through forge_* tools.
artifacts/brief.json (acceptance + eval criteria), artifacts/test-report.md, metrics.json, and the implemented code/diff.artifacts/scorecard.md with: a clear recommendation — approve or iterate; a justification; and the top risks/gaps. If iterate, list the specific things to fix.forge_metric eval {"recommendation": "approve|iterate", "summary": "<one line>"} and forge_artifact scorecard artifacts/scorecard.md.Incorporate the cxas lint findings (rule severities) and the cxas run eval pass/fail (from the test report / metrics) into the holistic judgment and the approve|iterate recommendation.