Verify Multispec

verify-multispec workflow (v3)

Overview

2-phase sub-workflow mapping CLAUDE.md "Verify 阶段 — 关键发布 / 大重构 PR 升级 Agent Team Pattern C" onto harnessed runtime (Phase v3.0-3.4 W0.13e — D-04 Stage ④ Verify 7 sub + D-11 Agent Teams + Pattern A sub-workflow ship)。

Per-phase config loads from workflows/verify/multispec/workflow.yaml; phase 01 spawns 4 background teammate (code-review + gstack-review + gstack-cso + gstack-qa) with Agent(name, run_in_background=true) — the team forms implicitly on the FIRST spawn (CC 2.1.178+ has no create step / no create tool), teammates 互相 SendMessage 质询 findings 是否 真问题 (NOT fire-and-forget); phase 02 mandatory 按名请求每个 teammate shut down (bundled Agent Teams cleanup discipline — 团目录在 session 退出时自动清理,无独立 teardown 工具, 剩下的纪律是别把 teammate 落在运行态)。

Capability refs

Sister workflows/capabilities.yaml entries:

• agent-teams-create — Bucket 5 Agent Teams (impl: claude-platform, cmd: Agent(name, run_in_background=true))
• agent-teams-send-message — Bucket 5 Agent Teams (impl: claude-platform, cmd: SendMessage)
• agent-teams-shutdown — Bucket 5 Agent Teams (impl: claude-platform, cmd: ask the <teammate-name> teammate to shut down)
• code-review — Bucket 1 mattpocock (teammate 1)
• gstack-review — Bucket 3 治理关卡 (teammate 2 Paranoid Staff Engineer)
• gstack-cso — Bucket 3 治理关卡 (teammate 3 安全审查)
• gstack-qa — Bucket 3 治理关卡 (teammate 4 端到端 QA)

Parallelism + on gate refs

Sister workflows/judgments/parallelism-gate.yaml:

• agent-teams-upgrade.fires — 5 OR-chain (teammate_send_message_needed / subagent_context_overflow / shared_task_list / opposing_hypothesis_debate / fullstack_three_way)

Phase-level on clause (critical-release 升级触发):

• if: phase.is_major_release == true or phase.is_large_refactor == true → action: invoke
• else → action: skip

Routing rules (bundled Agent Teams routing — workflows/judgments/parallelism-gate.yaml)

• ✅ 触发: 关键发布 / 大重构 PR (≥3 specialist 需互相质询而非 fire-and-forget)
• ❌ 跳过: 常规 PR / 单点任务 (sister verify-code-review fan-out + verify-paranoid 已够用且省 token)
• Token 估算 prereq: team_cost < 2 × subagent_cost (engine-level check; bundled cost guideline)
• Cleanup mandatory: phase 02-team-cleanup agent-teams-shutdown 必跑 (bundled cleanup discipline)

How to invoke

!harnessed checkpoint intent verify-multispec

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-multispec — that is the CI/headless path (in-process SDK spawn that blocks the session inside Claude Code).

1. Bash: harnessed prompt verify-multispec --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-multispec --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-multispec --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-multispec --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-multispec --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

• D-04 Stage ④ Verify 7 sub 分解
• D-11 Agent Teams 4-specialist Pattern C upgrade
• workflows/capabilities.yaml — agent-teams-{create,send-message,shutdown} + 4 specialist
• workflows/judgments/stage-routing.yaml — verify-multispec-critical-release trigger
• workflows/judgments/parallelism-gate.yaml — agent-teams-upgrade.fires (5 OR-chain)
• workflows/verify-work/workflow.yaml v2 SHIPPED phase 09-agent-team-multispecialist sister verbatim