Verify Second Opinion

verify-second-opinion workflow (v3)

Overview

1-phase sub-workflow — 编排契约改动时的跨模型第二意见 (Phase 54 T2)。

判据不是交给模型的判断题。harnessed facts 把 git diff --name-only <last release tag> 与 ORCHESTRATION SURFACE (judgments / capabilities / 事实 schema / resolver / exprBuilder / ledger / disciplines / 任一 workflows SKILL.md) 求交集,机械得出 requires_second_opinion。

基准是上一个 release tag,不是 merge-base origin/main —— 后者在 commit-即-push-main 的纪律下塌缩成 HEAD,只看得见未提交改动,会漏掉多-commit milestone 的早期 commit。

算不出来 (无 tag / 无 git / diff 失败) 时取 false,并在每轮 <workflow-state> 断点打出 SECOND-OPINION: 判据不可用(<原因>)。不乱 fire,且不静默。

Capability refs

Sister workflows/capabilities.yaml entries:

• codex — 第二意见 / cross-AI peer review via codex (impl: gstack, cmd: /codex)

Gate ref

Sister workflows/judgments/stage-routing.yaml:

• verify-second-opinion-orchestration.fires — phase.stage == 'verify' and requires_second_opinion == true

Routing rules

• ✅ 触发: 本次改动触及编排契约本身 (judgments / capabilities.yaml / phaseFactContext / facts.ts / judgmentResolver / exprBuilder / ledger.ts / disciplines / 任一 SKILL.md)
• ❌ 跳过: 纯 installer / 文档 / 测试 / 依赖 bump —— 改这些不会改变引擎的决策方式

交付物即证据。 本 sub fire 了就必须交 second-opinion.md;不交则 checkpoint complete fail-closed 阻断,--force 在 ledger 记 evidence_status: overridden。跳过因此是台账上 一行,不是隐形的。

How to invoke

!harnessed checkpoint intent verify-second-opinion

The banner above (when present) means this invocation is REGISTERED with the engine (an intent marker) — not yet compliant: the steps below (prompt → spawn → checkpoint complete) resolve it, and a per-turn <workflow-intent> reminder persists until they run.

The numbered sequence below is the state machine — execute it with Bash. Do NOT improvise an equivalent flow from the Overview above: freelancing bypasses the engine (no ledger, no evidence guard). harnessed gives you the spawn-ready prompt; YOU spawn the subagent with a CC-native Task / Agent tool (keeps the session responsive + lets clarification round-trips reach the user).

Do NOT pipe to harnessed run verify-second-opinion — that is the CI/headless path (in-process SDK spawn that blocks the session inside Claude Code).

1. Bash: harnessed prompt verify-second-opinion --task "$ARGUMENTS" --json → parse {prompt, max_iterations, model}.
2. Spawn a CC-native subagent (Task / Agent tool) with that prompt and model, then drive delivery with harnessed's own completion gate:
   • on return, write the subagent's final output to a file and run harnessed checkpoint complete verify-second-opinion --result-file <path> — it is fail-closed on the declared artifacts, the TDD boundary, and the verbatim <promise>COMPLETE</promise>.
   • if it blocks, run harnessed checkpoint fail verify-second-opinion --failing-tests <n> to record the attempt; it prints BUDGET-EXHAUSTED / NO-PROGRESS / BREAK-LOOP when a stop condition is reached.
   • respawn ONLY while none of those three has fired. Any one of them means stop: re-scope the subtask, fix the blocker, or escalate to the user. Never respawn past a stop directive.
3. If the output contains STATUS: NEEDS_CLARIFICATION + a question list: STOP, relay them verbatim via AskUserQuestion, append the answers to the spec, then re-spawn the same sub.
4. On <promise>COMPLETE</promise>: write the subagent’s final output to a file, then Bash harnessed checkpoint complete verify-second-opinion --result-file <path> --summary "<one-line>". Fail-CLOSED — it blocks unless every declared artifacts_expected file exists, the TDD boundary passes (non-empty evidence / both the red and green sides present / the test file was not deleted), and the result carries a verbatim <promise>COMPLETE</promise> (or a structured COMPLETE status). --result <text> is the inline variant; --result-file wins and is quoting-safe on Windows. --force records an audited override (evidence_status=overridden) — it does not silently pass.
5. If the complete gate blocked: Bash harnessed checkpoint fail verify-second-opinion --failing-tests <n> to record the attempt. It prints BUDGET-EXHAUSTED / NO-PROGRESS / BREAK-LOOP once a stop condition is reached. Respawn ONLY while none of those three has fired; any one of them means STOP — re-scope the subtask, fix the blocker, or escalate to the user.

References

• Phase 54 SPEC — .planning/phases/54-declaration-evaluation-parity/SPEC.md
• workflows/capabilities.yaml — codex
• workflows/judgments/stage-routing.yaml — verify-second-opinion-orchestration trigger
• workflows/defaults.yaml — ralph_max_iterations.verify-second-opinion.*
• src/cli/facts.ts — deriveSecondOpinion / ORCHESTRATION_SURFACE