@loomhelm/multi-task-flow 0.0.0-stage → 0.2.1

package/toolkit/skills/dev-plan/references/plan-cases/positive-plan-constraints.md

@@ -0,0 +1,45 @@
+# Plan constraint raw-plan fixture
+
+## Summary
+
+- 交付一个可直接创建并收敛 PR 的 Fast-track 变更。
+
+## Flow Decision
+
+- `flow_type`: `fast-track`
+- `flow_source`: `user_explicit`
+- `lock_status`: `locked`
+- `empirical_acceptance`: `not_required`
+- `empirical_acceptance_rationale`: `范围合同可由确定性本地检查证明，不依赖外部行为观察。`
+- `review_tier`: `Tier 3`
+- `review_rationale`: `当前 PR candidate 需要 direct review evidence。`
+- `runtime_review_strategy`: `readiness 和验证完成后运行时创建 required lanes。`
+- `review_profile_recommendation`: `四至六个 formal lane 建议使用 gpt-6-luna + high pair。`
+- 开发基线：`origin/main`
+- 收口条件：`merge-ready / Step 5C Ready`
+
+## Goals
+
+- G1: 收敛当前 PR。创建或复用 PR，并把 current head 的 CI/review 收敛到 Step 5C。
+
+## Constraints
+
+- C1: 远端验收只使用当前 head 的受控输入。 | applies_to=`A1`
+
+## Non-goals
+
+- N1: 不 push debug logs，也不合并 PR 或执行本地 cleanup。
+
+## Task Orchestration
+
+- wave: 1
+  - main-agent => 创建或复用 PR 并收敛 current head (skill: $fast-flow)
+
+## Acceptance & Validation
+
+- A1 [contract/direct] current-head PR convergence | scope=`plan` | phase=`pre_merge` | evidence=`test:raw-plan` | assertions=`live PR exists,current head CI/review converge,Step 5C Ready`
+
+## Assumptions/Defaults
+
+- Runtime release behavior is derived from target repository policy and current PR facts; the Agent default decision card reports publish applicability, components, channel, bump rule, and predicted version, shows the policy source, and requires owner confirmation including the version expectation; missing/conflicting inputs remain unknown/blocked. 已确定。
+- PR 的远端写入由已锁定的 Fast-track 授权：已确定。

package/toolkit/skills/dev-plan/references/plan-cases/positive-plan-task-metadata.md

@@ -0,0 +1,40 @@
+# Plan task metadata raw-plan fixture
+
+## Summary
+
+- 交付带普通任务元数据的范围合同。
+
+## Flow Decision
+
+- `flow_type`: `normal`
+- `flow_source`: `user_explicit`
+- `lock_status`: `locked`
+- `empirical_acceptance`: `not_required`
+- `empirical_acceptance_rationale`: `该格式合同只由确定性本地检查证明。`
+- `review_tier`: `Tier 1`
+- `review_rationale`: `本计划只校验任务格式，不需要 formal review gate。`
+- `runtime_review_strategy`: `Tier 1 不启动 formal review；仅执行计划合同校验。`
+- `review_profile_recommendation`: `Tier 1 不需要 formal lane profile；如需 sidecar 由运行时单独选择。`
+- 开发基线：`origin/main`
+
+## Goals
+
+- G1: 建立范围合同。交付受限的计划输出格式。
+
+## Non-goals
+
+- N1: 不修改外部项目文档。
+
+## Task Orchestration
+
+- wave: 1
+  - main-agent => 为任务对象增加 owner: "frontend" 元数据，并以 adr: 记录架构决定 (skill: $normal-flow)
+
+## Acceptance & Validation
+
+- A1 [contract/direct] plan format | scope=`plan` | phase=`pre_merge` | evidence=`test:format` | assertions=`task metadata remains valid`
+
+## Assumptions/Defaults
+
+- Runtime release behavior is derived from target repository policy and current PR facts; the Agent default decision card reports publish applicability, components, channel, bump rule, and predicted version, shows the policy source, and requires owner confirmation including the version expectation; missing/conflicting inputs remain unknown/blocked. 已确定。
+- 本地 checker 可用：已确定。

package/toolkit/skills/dev-plan/references/plan-cases/positive-wave-thread-v2.md

@@ -0,0 +1,57 @@
+# Wave Thread v2 batch plan raw-plan fixture
+
+## Summary
+
+- 固定一个多波次父子批次和一个直接批次，验证四层交付拓扑及阶段验收绑定。
+- 本 raw-plan fixture 给出 A1-A10 的离线 scope/phase/locator 映射；A10 的 live PR、CI、formal review 和 Step 5C 事实仍由 delivery gate 在候选 PR 上单独证明。
+
+## Flow Decision
+
+- `flow_type`: `wave-thread`
+- `flow_source`: `user_explicit`
+- `lock_status`: `locked`
+- `empirical_acceptance`: `not_required`
+- `empirical_acceptance_rationale`: `批次、波次、任务和验收绑定由确定性本地契约与离线回放证明。`
+- `review_tier`: `Tier 3`
+- `review_rationale`: `v2 状态、事件和兼容适配需要当前候选的完整 formal review。`
+- `runtime_review_strategy`: `Review Readiness 通过后创建 required formal lanes。`
+- `review_profile_recommendation`: `四个 formal lane 建议使用 gpt-5.6-luna + xhigh，不预设审查者身份。`
+- 开发基线：`origin/main`
+- 收口条件：`merge-ready / Step 5C Ready`
+
+## Goals
+
+- G1: 固定父子批次的多波次编排和直接批次的单任务最终 PR。每个批次声明自己的观察期限、依赖、条件和验收证据。
+
+## Non-goals
+
+- N1: 不创建真实任务、PR、发布或部署资源。
+
+## Task Orchestration
+
+- batch:batch-a [parent-child] => target=integration/main;depends_on=none;condition=true;acceptance=A1,A6,A7,A8,A9;observation_window=2026-12-31T00:00:00Z;parent_branch=integration/main;child_pr=required;topology=sibling-prs;event_protocol=wave-thread/v2;release_policy=branch-preview
+  - wave: 1
+    - task:task-a1 [gpt-6-luna, max] => owned_surfaces=src/a;depends_on=none;condition=true;acceptance=A2;skill=$wave-thread-flow;title=Prepare canonical facts;purpose=Validate the canonical batch and task facts before projection.;outputs=validated parser facts for the first wave
+  - wave: 2
+    - task:task-a2 [gpt-5.6-luna, xhigh] => owned_surfaces=src/b;depends_on=task-a1/integrated;condition=true;acceptance=A3;skill=$wave-thread-flow;title=Project the review view;purpose=Render the approved owner-facing view from validated facts.;outputs=transient Markdown projection
+- batch:batch-b [direct] => target=main;depends_on=none;condition=true;acceptance=A4,A10;observation_window=2026-12-31T00:00:00Z;parent_branch=main;child_pr=none;topology=direct-final-pr;event_protocol=wave-thread/v2;release_policy=branch-preview
+  - wave: 1
+    - task:task-b1 [gpt-5.6-luna, max] => owned_surfaces=src/c;depends_on=none;condition=true;acceptance=A5;skill=$wave-thread-flow;title=Verify delivery boundaries;purpose=Confirm direct-batch delivery and candidate isolation.;outputs=delivery validation evidence
+
+## Acceptance & Validation
+
+- A1 [contract/direct] batch topology | scope=`batch:batch-a` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_default_parent_child_batch_can_have_multiple_waves` | assertions=`one parent PR across multiple waves`
+- A2 [behavior/direct] first task | scope=`task:task-a1` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_multi_wave_batch_can_advance_before_future_wave_tasks_exist` | assertions=`task integration and archive precede dependent wave`
+- A3 [behavior/direct] dependent task | scope=`task:task-a2` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_multi_wave_batch_can_advance_before_future_wave_tasks_exist` | assertions=`later wave waits for integrated task`
+- A4 [contract/direct] direct topology | scope=`batch:batch-b` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_direct_mode_implementation_pr_is_final_pr` | assertions=`implementation PR is final PR`
+- A5 [behavior/direct] direct task | scope=`task:task-b1` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_direct_mode_implementation_pr_is_final_pr` | assertions=`direct readiness does not archive itself`
+- A6 [permission/direct] closeout boundary | scope=`batch:batch-a` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_parent_child_finalize_requires_conversation_archive_receipt` | assertions=`parent-child closeout requires native archive receipt`
+- A7 [migration/direct] v1 compatibility | scope=`batch:batch-a` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_v1_upgrade_is_read_only_and_preserves_default_batch` | assertions=`legacy state maps to one default batch without mutating input`
+- A8 [behavior/direct] completion timeout | scope=`batch:batch-a` | phase=`post_deploy` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_observation_deadline_records_timeout_without_resetting_deadline` | assertions=`timeout preserves absolute deadline and blocks dispatch`
+- A9 [mechanical/direct] replay quality | scope=`batch:batch-a` | phase=`pre_merge` | evidence=`skills/wave-thread-flow/scripts/wave_thread_replay_check.ts::check_v2_recovery` | assertions=`offline v2 event protocol deferred delivery merge gate and recovery replay pass`
+- A10 [contract/direct] final delivery | scope=`batch:batch-b` | phase=`pre_merge` | evidence=`test_wave_thread_v2.ts::WaveThreadV2Tests.test_direct_mode_implementation_pr_is_final_PR` | assertions=`offline direct topology is proven; live PR CI review and Step 5C remain delivery-gated`
+
+## Assumptions/Defaults
+
+- Runtime release behavior is derived from target repository policy and current PR facts; the Agent default decision card reports publish applicability, components, channel, bump rule, and predicted version, shows the policy source, and requires owner confirmation including the version expectation; missing/conflicting inputs remain unknown/blocked. 已确定。
+- 本 fixture 只证明本地 parser、状态机和离线 replay；不宣称原生 Codex 事件往返成功。已确定。

package/toolkit/skills/dev-plan/references/replay-cases/CASES.txt

@@ -0,0 +1,248 @@
+# expect  path  required_code
+fail historical-flow-not-locked-before-plan.jsonl FLOW_NOT_LOCKED_BEFORE_PLAN
+fail historical-missing-flow-type.jsonl MISSING_FLOW_TYPE
+fail missing-task-orchestration.jsonl MISSING_TASK_ORCHESTRATION_SECTION
+fail invalid-task-orchestration-content.jsonl TASK_ORCHESTRATION_NON_LIST_CONTENT
+fail flat-task-orchestration-list.jsonl TASK_ORCHESTRATION_TASK_OUTSIDE_WAVE
+fail checkbox-task-orchestration.jsonl TASK_ORCHESTRATION_CHECKBOX_FORBIDDEN
+fail missing-task-orchestration-task.jsonl TASK_ORCHESTRATION_EMPTY_WAVE
+fail wave-sequence-jump.jsonl TASK_ORCHESTRATION_WAVE_SEQUENCE_INVALID
+fail wave-sequence-repeat.jsonl TASK_ORCHESTRATION_WAVE_SEQUENCE_INVALID
+fail duplicate-task-in-wave.jsonl TASK_ORCHESTRATION_DUPLICATE_TASK_IN_WAVE
+fail missing-task-skill-binding.jsonl TASK_ORCHESTRATION_MISSING_SKILL_BINDING
+fail summary-blank-bullet.jsonl SUMMARY_EMPTY
+fail missing-goals.jsonl MISSING_GOALS_SECTION
+fail missing-non-goals.jsonl MISSING_NON_GOALS_SECTION
+fail goals-empty-orchestration-only.jsonl GOALS_EMPTY
+fail goal-blank-bullet.jsonl GOAL_TITLE_EMPTY
+fail goal-without-detail.jsonl GOAL_DETAIL_MISSING
+fail non-goals-blank-bullet.jsonl NON_GOAL_EMPTY
+fail invalid-section-order.jsonl SECTION_ORDER_INVALID
+fail persisted-plan-artifact-in-orchestration.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail mixed-plan-cleanup-create-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail decision-log-artifact-in-orchestration.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail handoff-log-artifact-in-orchestration.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail decision-log-future-work-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail handoff-log-future-work-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail decision-log-timeline-planning-handoff-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail handoff-log-history-agent-session-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail plan-md-handoff-job-future-sessions-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail decision-log-job-future-handoffs-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail generate-handoff-log-future-sessions-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail keep-handoff-log-future-sessions-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail decision-log-workflow-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail handoff-log-pipeline-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail move-plan-md-handoff-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail prepare-handoff-log-next-session-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail create-plan-md-implementation-steps-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail update-plan-md-implementation-steps-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail read-plan-md-handoff-context-artifact.jsonl TASK_ORCHESTRATION_PERSISTED_PLAN_ARTIFACT_FORBIDDEN
+fail placeholder-acceptance.jsonl ACCEPTANCE_PLACEHOLDER
+fail blank-acceptance-bullet.jsonl ACCEPTANCE_EMPTY
+fail acceptance-notes-only.jsonl ACCEPTANCE_PLACEHOLDER
+fail acceptance-follow-up-items-only.jsonl ACCEPTANCE_PLACEHOLDER
+fail acceptance-tba-only.jsonl ACCEPTANCE_PLACEHOLDER
+fail goal-placeholder-detail.jsonl GOAL_DETAIL_MISSING
+fail goal-tba-detail.jsonl GOAL_DETAIL_MISSING
+fail non-goals-placeholder.jsonl NON_GOAL_PLACEHOLDER
+fail non-goals-none-placeholder.jsonl NON_GOAL_PLACEHOLDER
+fail redundant-execution-steps.jsonl REDUNDANT_EXECUTION_STEPS_SECTION
+fail missing-flow-question.jsonl FLOW_QUESTION_MISSING
+fail unrelated-flow-wording.jsonl FLOW_QUESTION_MISSING
+fail unrelated-flow-wording-default-fast-track.jsonl DEFAULT_FAST_TRACK_NOT_ALLOWED_HERE
+fail default-fast-track-missing-merge-ready-stop.jsonl DEFAULT_FAST_TRACK_MISSING_MERGE_READY_STOP
+fail flow-decision-mismatch-lock.jsonl FLOW_DECISION_MISMATCH_LOCK
+fail conflicting-flow-signals-default-fast-track.jsonl DEFAULT_FAST_TRACK_NOT_ALLOWED_HERE
+fail skill-binding-no-dollar-prefix.jsonl TASK_ORCHESTRATION_SKILL_REF_FORMAT_INVALID
+fail skill-binding-mixed-dollar-and-plain.jsonl TASK_ORCHESTRATION_SKILL_REF_FORMAT_INVALID
+fail invalid-flow-type-substring.jsonl UNKNOWN_FLOW_TYPE
+fail invalid-user-flow-lock-substring.jsonl FLOW_NOT_LOCKED_BEFORE_PLAN
+fail implement-only-forbidden-task-orchestration.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-sync-task-orchestration.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-copy-spec-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-create-for-tests-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-update-tests-slug-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-sync-against-existing-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-update-existing-spec-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-create-migrated-copy-task.jsonl IMPLEMENT_ONLY_FORBIDDEN_TASK_IN_ORCHESTRATION
+fail implement-only-forbidden-skill-topic-spec.jsonl IMPLEMENT_ONLY_FORBIDDEN_SKILL
+fail implement-only-forbidden-skill-normal-flow.jsonl IMPLEMENT_ONLY_FORBIDDEN_SKILL
+fail missing-development-baseline.jsonl MISSING_DEVELOPMENT_BASELINE
+fail flow-decision-baseline-only-outside-section.jsonl MISSING_DEVELOPMENT_BASELINE
+fail abstract-development-baseline.jsonl ABSTRACT_DEVELOPMENT_BASELINE
+fail abstract-development-baseline-combined-phrase.jsonl ABSTRACT_DEVELOPMENT_BASELINE
+fail filesystem-path-development-baseline.jsonl INVALID_DEVELOPMENT_BASELINE
+fail context-dependent-alias-development-baseline.jsonl INVALID_DEVELOPMENT_BASELINE
+fail context-dependent-remote-head-development-baseline.jsonl INVALID_DEVELOPMENT_BASELINE
+fail placeholder-development-baseline.jsonl INVALID_DEVELOPMENT_BASELINE
+fail explicit-owner-baseline-mismatch.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-after-flow-answer.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-after-standalone-flow-restatement.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-plain-chinese.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-yong-plain-chinese.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-directional-baseline.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-base-ref-equals.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-base-equals.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-english-is.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-legal-trailing-punct.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-short-ref.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-legal-trailing-semicolon-plain.jsonl OWNER_BASELINE_MISMATCH
+fail explicit-owner-baseline-mismatch-plan-page-upload-api-scope-clarification.jsonl OWNER_BASELINE_MISMATCH
+pass positive-fast-track.jsonl -
+pass ../plan-cases/positive-fast-track.md - raw-plan
+pass ../plan-cases/positive-fast-track-aggregate-pr-exclusion.md - raw-plan
+pass ../plan-cases/positive-empirical-acceptance.md - raw-plan
+pass ../plan-cases/positive-phase-aware-direct.md - raw-plan
+fail ../plan-cases/positive-multi-task-flow.md DEV_PLAN_MULTI_TASK_FLOW_FORBIDDEN raw-plan
+fail ../plan-cases/negative-multi-task-flow-single-task.md DEV_PLAN_MULTI_TASK_FLOW_FORBIDDEN raw-plan
+fail ../plan-cases/negative-empirical-acceptance-missing.md EMPIRICAL_ACCEPTANCE_DISPOSITION_REQUIRED raw-plan
+fail ../plan-cases/negative-empirical-acceptance-scenarios.md EMPIRICAL_ACCEPTANCE_ASSERTIONS_INVALID raw-plan
+fail ../plan-cases/negative-empirical-acceptance-placeholders.md EMPIRICAL_ACCEPTANCE_ASSERTIONS_INVALID raw-plan
+fail ../plan-cases/negative-direct-review-tier-missing.md FORMAL_REVIEW_TIER_REQUIRED raw-plan
+fail ../plan-cases/negative-fast-track-pr-forbidden.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-no-pr-local.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-avoid-pr-delivery.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-avoid-pr-creation.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-avoid-open-pr.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-no-direct-pr.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-direct-pr-omitted.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-local-only-delivery.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-no-pr-delivery.md FAST_TRACK_PR_FORBIDDEN raw-plan
+fail ../plan-cases/negative-fast-track-pr-target-only.md FAST_TRACK_PR_ACCEPTANCE_MISSING raw-plan
+fail ../plan-cases/negative-fast-track-pr-acceptance-missing.md FAST_TRACK_PR_ACCEPTANCE_MISSING raw-plan
+fail ../plan-cases/negative-fast-track-pr-acceptance-split.md FAST_TRACK_PR_ACCEPTANCE_MISSING raw-plan
+fail ../plan-cases/negative-fast-track-unlocked.md LOCK_STATUS_NOT_LOCKED raw-plan
+fail ../plan-cases/negative-phase-aware-missing-phase.md ACCEPTANCE_MAPPING_INVALID raw-plan
+pass positive-agent-profiles.jsonl -
+fail ../plan-cases/positive-initiative-bootstrap-v5.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/positive-initiative-delivery-v5.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-missing.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-trailing-comma.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-duplicate.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-unsorted.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-duplicate-path.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-initiative-adr-contracts-malformed-sha.md RETIRED_FLOW_TYPE raw-plan
+pass ../plan-cases/positive-plan-constraints.md - raw-plan
+pass ../plan-cases/positive-plan-task-metadata.md - raw-plan
+fail ../plan-cases/positive-wave-thread-v2.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-wave-thread-v2-duplicate-contract.md RETIRED_FLOW_TYPE raw-plan
+fail positive-wave-thread-v2-legacy.jsonl RETIRED_FLOW_TYPE
+fail ../plan-cases/negative-plan-provenance-section.md PLAN_PROVENANCE_SECTION_FORBIDDEN raw-plan
+fail ../plan-cases/negative-plan-provenance-variant.md PLAN_PROVENANCE_SECTION_FORBIDDEN raw-plan
+fail ../plan-cases/negative-plan-constraint-acceptance.md PLAN_CONSTRAINT_ACCEPTANCE_UNKNOWN raw-plan
+fail ../plan-cases/negative-plan-constraint-task-item.md TASK_ORCHESTRATION_CONSTRAINT_REFERENCE_FORBIDDEN raw-plan
+pass positive-subagent-readonly-negation.jsonl -
+pass positive-subagent-indirect-readonly-negation.jsonl -
+pass positive-subagent-git-readonly-commands.jsonl -
+pass positive-subagent-git-config-worktree-readonly.jsonl -
+fail negative-agent-profile-missing.jsonl TASK_ORCHESTRATION_AGENT_PROFILE_MISSING
+fail negative-agent-profile-invalid.jsonl TASK_ORCHESTRATION_AGENT_PROFILE_INVALID
+fail negative-main-agent-profile.jsonl MAIN_AGENT_PROFILE_FORBIDDEN
+fail negative-legacy-profile-text.jsonl TASK_ORCHESTRATION_LEGACY_PROFILE_FORBIDDEN
+fail negative-specialist-review-profile.jsonl FORMAL_REVIEW_PREDECLARATION_FORBIDDEN
+fail negative-specialist-review-effort.jsonl FORMAL_REVIEW_PREDECLARATION_FORBIDDEN
+fail negative-formal-main-agent-predeclaration.jsonl FORMAL_REVIEW_PREDECLARATION_FORBIDDEN
+fail negative-formal-evidence-predeclaration.jsonl FORMAL_REVIEW_PREDECLARATION_FORBIDDEN
+fail ../plan-cases/negative-runtime-strategy-ungated.md FORMAL_REVIEW_RUNTIME_STRATEGY_INVALID raw-plan
+fail ../plan-cases/negative-runtime-profile-pair.md FORMAL_REVIEW_PROFILE_RECOMMENDATION_INVALID raw-plan
+fail ../plan-cases/negative-runtime-profile-reviewer-identity.md FORMAL_REVIEW_PROFILE_RECOMMENDATION_INVALID raw-plan
+pass positive-runtime-formal-review.jsonl -
+fail negative-task-executor-noninitiative.jsonl TASK_EXECUTOR_FORBIDDEN_FLOW
+fail negative-subagent-shared-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-shared-write-english.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-local-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-contradictory-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-generic-english-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-filesystem-write-actions.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-implementing-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-generic-implementation-write.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-implementation-multiword.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-git-write-commands.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-git-and-shell-write-commands.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-git-global-options.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-subagent-git-config-worktree.jsonl SUB_AGENT_SHARED_WRITE_FORBIDDEN
+fail negative-task-agent-alias.jsonl TASK_ORCHESTRATION_INVALID_ITEM
+pass positive-large-coupled-default-fast-track.jsonl -
+fail negative-default-initiative-eligibility-unproven.jsonl DEFAULT_INITIATIVE_ELIGIBILITY_UNPROVEN
+fail positive-default-initiative-from-planned-scope.jsonl DEFAULT_INITIATIVE_INVALID_FLOW_DECISION
+fail negative-initiative-prerequisite-silent-downgrade.jsonl INITIATIVE_PREREQUISITES_UNAVAILABLE
+pass positive-subagent-review-implementation.jsonl -
+pass positive-subagent-git-stash-readonly.jsonl -
+fail negative-subagent-git-readonly-mixed-commands.jsonl -
+fail negative-new-delivery-topology.jsonl LEGACY_ROUTING_FIELD_FORBIDDEN
+fail negative-new-mixed-top-level-flow.jsonl UNKNOWN_FLOW_TYPE
+fail negative-initiative-with-legacy-fast-alias.jsonl UNKNOWN_FLOW_TYPE
+pass positive-no-delivery-topology.jsonl -
+pass positive-direct-dev-plan-issue-parser.jsonl -
+pass positive-normal.jsonl -
+pass positive-normal-decision-log-entries.jsonl -
+pass positive-normal-decision-log-ui.jsonl -
+pass positive-normal-handoff-log-ui.jsonl -
+pass positive-normal-plan-md-cleanup.jsonl -
+pass positive-normal-next-session-decision-log-api.jsonl -
+pass positive-normal-future-handoff-log-timeline.jsonl -
+pass positive-normal-plan-file-upload-api.jsonl -
+pass positive-normal-plan-md-user-upload.jsonl -
+pass positive-normal-plan-md-importer.jsonl -
+pass positive-normal-plan-md-import-review-page.jsonl -
+pass positive-normal-plan-md-planning-ui-import.jsonl -
+pass positive-normal-agent-decision-log-api.jsonl -
+pass positive-normal-agent-handoff-log-exporter.jsonl -
+pass positive-normal-next-session-decision-log-exporter.jsonl -
+pass positive-normal-plan-template-decision-log-history.jsonl -
+pass positive-normal-plan-template-handoff-log-history.jsonl -
+pass positive-normal-progress-log-ui.jsonl -
+pass positive-default-fast-track-no-flow.jsonl -
+pass positive-flow-lock-direct-local-merge-skill.jsonl -
+pass positive-flow-lock-standalone-normal-punctuation.jsonl -
+pass positive-normal-pending-state-acceptance.jsonl -
+pass positive-flow-lock-rui.jsonl -
+pass positive-flow-lock-latest-explicit-user.jsonl -
+pass positive-flow-lock-fast-flow-keyed.jsonl -
+pass non-baseline-based-on-docs-text.jsonl -
+pass positive-explicit-english-branch-baseline-is.jsonl -
+pass positive-explicit-branch-baseline-after-flow-answer.jsonl -
+pass positive-explicit-branch-baseline-after-standalone-flow-restatement.jsonl -
+pass positive-explicit-branch-baseline-directional-baseline.jsonl -
+pass positive-explicit-branch-baseline-base-ref-equals.jsonl -
+pass positive-explicit-branch-baseline.jsonl -
+pass positive-explicit-branch-baseline-legal-trailing-punct.jsonl -
+pass positive-explicit-branch-baseline-legal-trailing-punct-comma-clause.jsonl -
+pass positive-explicit-branch-baseline-legal-trailing-semicolon-plain.jsonl -
+pass positive-directional-product-wording-no-baseline.jsonl -
+pass positive-explicit-branch-baseline-plain-chinese.jsonl -
+pass positive-explicit-branch-baseline-yong-plain-chinese.jsonl -
+pass positive-explicit-branch-baseline-trailing-comma-clause.jsonl -
+pass positive-explicit-branch-baseline-special-chars.jsonl -
+pass positive-explicit-branch-baseline-feature-head.jsonl -
+pass positive-explicit-branch-baseline-topic-prefix.jsonl -
+pass positive-explicit-branch-baseline-short-ref.jsonl -
+pass positive-explicit-branch-baseline-replace-clause.jsonl -
+pass positive-explicit-commit-baseline.jsonl -
+pass positive-default-mainline-concrete-baseline.jsonl -
+pass positive-default-mainline-token-ref.jsonl -
+pass positive-explicit-branch-baseline-placeholder-token-ref.jsonl -
+pass positive-explicit-branch-baseline-doc-like-ref.jsonl -
+pass positive-multi-plan-baseline-reset-generic-reask.jsonl -
+pass positive-multi-plan-baseline-reset-merge-ready.jsonl -
+pass positive-multi-plan-baseline-reset-plan-task-wording.jsonl -
+pass positive-multi-plan-baseline-reset-merge-queue-api.jsonl -
+pass positive-multi-plan-baseline-reset.jsonl -
+pass positive-multi-plan-baseline-reset-no-dev-plan-token.jsonl -
+pass positive-multi-plan-baseline-reset-replan-task.jsonl -
+pass positive-multi-plan-baseline-reset-replan-merge-ready-clause.jsonl -
+pass positive-multi-plan-baseline-reset-explicit-dev-plan-with-continue-wording.jsonl -
+pass positive-equivalent-remote-ref-baseline.jsonl -
+pass positive-explicit-branch-baseline-after-prose-flow-answer.jsonl -
+pass positive-scope-clarification-plan-md-no-reset.jsonl -
+pass positive-scope-clarification-import-page-no-reset.jsonl -
+pass positive-scope-clarification-new-plan-file-upload-api-no-reset.jsonl -
+pass positive-scope-clarification-plan-page-upload-api-no-reset.jsonl -
+pass positive-skill-binding-fullwidth-comma.jsonl -
+pass positive-implement-only-pure.jsonl -
+pass positive-implement-only-add-tests-reference.jsonl -
+pass positive-implement-only-parser-reference.jsonl -
+pass positive-implement-only-update-parser-reference.jsonl -
+fail ambiguous-continue-fallback.jsonl FLOW_QUESTION_MISSING
+fail ../plan-cases/positive-parallel-pr.md RETIRED_FLOW_TYPE raw-plan
+fail ../plan-cases/negative-parallel-pr-overlap.md RETIRED_FLOW_TYPE raw-plan

package/toolkit/skills/dev-plan/references/replay-cases/README.md

@@ -0,0 +1,190 @@
+# Plan Contract Replay Cases
+
+- `CASES.txt` defines expected results for each replay input.
+- `raw-plan` validates current explicit `$dev-plan` Markdown fixtures without requiring conversation provenance in the plan body.
+- In these fixtures, `merge-ready / Step 5C Ready` wording inside `<proposed_plan>` is a planned stop-condition marker, not a live PR status claim.
+- Mandatory coverage:
+  - one repo-local historical failure fixture for `FLOW_NOT_LOCKED_BEFORE_PLAN`;
+  - one repo-local historical failure fixture for missing flow declaration in plan text;
+- retired Initiative and Parallel PR fixtures are retained as explicit rejection cases so the active checker does not advertise superseded routes as valid plans;
+  - one local failure fixture for missing `Task Orchestration` section;
+  - one local failure fixture for non-list content in `Task Orchestration`;
+  - one local failure fixture for task item outside wave;
+  - one local failure fixture for checkbox markers;
+  - one local failure fixture for empty wave (missing nested task item);
+  - one local failure fixture for wave jump;
+  - one local failure fixture for wave repeat;
+  - one local failure fixture for duplicate task item in the same wave;
+  - one local failure fixture for missing task-level skill binding;
+  - one local failure fixture for missing `Goals`;
+  - one local failure fixture for missing `Non-goals`;
+  - one local failure fixture for empty `Goals`;
+  - one local failure fixture for goal items without follow-up detail text;
+  - one local failure fixture for invalid section order (`Task Orchestration` before `Goals` / `Non-goals`);
+  - one local failure fixture for persisted plan artifacts in `Task Orchestration`;
+  - one local failure fixture for mixed cleanup/create `PLAN.md` wording that must still be rejected as a persisted plan artifact;
+  - one local failure fixture for plain planning `decision log` artifacts in `Task Orchestration`;
+  - one local failure fixture for planning `handoff log` artifacts in `Task Orchestration`;
+  - one local failure fixture proving that generic `PLAN.md` handoff job wording with future-session context is still rejected;
+  - one local failure fixture proving that generic `decision log job` wording with future-handoff context is still rejected;
+  - one local failure fixture proving that `generate` still counts as a persisted-artifact action for planning handoff logs;
+  - one local failure fixture proving that `keep` still counts as a persisted-artifact action for planning handoff logs;
+  - one local failure fixture proving that generic `decision log workflow` wording is still rejected as a planning artifact;
+  - one local failure fixture proving that generic `handoff log pipeline` wording is still rejected as a planning artifact;
+  - one local failure fixture proving that moving a generated `PLAN.md` into a future-handoff location is still rejected as a persisted plan artifact;
+  - one local failure fixture proving that `prepare handoff log for next session` wording is still rejected as a planning artifact;
+  - one local failure fixture proving that generic `create PLAN.md file with implementation steps` wording is still rejected as a persisted plan artifact;
+  - one local failure fixture proving that generic `update PLAN.md with implementation steps` wording is still rejected as a persisted plan artifact;
+  - one local failure fixture for placeholder-only `Acceptance & Validation` text;
+  - one local failure fixture for `Acceptance & Validation` containing only `Notes`;
+  - one local failure fixture for `Acceptance & Validation` containing only `Follow-up items`;
+  - one local failure fixture for `Acceptance & Validation` containing only `TBA`;
+  - one local failure fixture for goal detail lines containing only placeholder text such as `TBD`;
+  - one local failure fixture for goal detail lines containing only placeholder text such as `TBA`;
+  - one local failure fixture for `Non-goals` containing only placeholder bullets;
+  - one positive fixture for a `C*` Constraint bound to a concrete acceptance item;
+  - failure fixtures for provenance leakage variants, a Constraint targeting an unknown acceptance item, and a Task Orchestration item that references a declared Constraint; ordinary task metadata that happens to use `owner` or `adr` remains valid.
+  - one positive normal-flow fixture proving that product-facing `decision log UI` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `workflow decision log entries` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `handoff log UI/API` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that cleanup tasks removing legacy `PLAN.md` references are not misclassified as persisted plan artifacts;
+  - one positive normal-flow fixture proving that product-facing `next-session decision log entries API` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `future handoff log timeline` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that generic product `plan file upload API` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product importer wording around customer-provided `PLAN.md` is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `agent decision log export API` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `next-session decision log` exporter wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `plan template decision log history` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `plan template handoff log history` wording is not misclassified as a persisted plan artifact;
+  - one positive normal-flow fixture proving that product-facing `progress log UI/API` wording is not misclassified as a persisted plan artifact;
+  - one local failure fixture for redundant `Execution Steps` section;
+  - one local failure fixture for missing `flow_type` question before plan output;
+  - one local failure fixture for unrelated flow wording (non-`flow_type` question);
+  - one local failure fixture proving flow-type/routing wording in the user request must also block silent `default_fast_track` when it does not actually lock a flow;
+  - one local failure fixture proving `default_fast_track` plans must still name a `merge-ready / Step 5C Ready` stop condition instead of stopping at plain PR creation;
+  - one local failure fixture for locked `flow_type` mismatch against `Flow Decision.flow_type`;
+  - one local failure fixture proving mutually conflicting flow signals must not silently fall back to `default_fast_track`;
+  - one local failure fixture for skill binding without `$` prefix in `Task Orchestration`;
+  - one local failure fixture for mixed skill refs where at least one ref misses `$` prefix in `Task Orchestration`;
+  - one local failure fixture for malformed `Flow Decision.flow_type` substring value (must report `UNKNOWN_FLOW_TYPE`);
+  - one local failure fixture for malformed user explicit `flow_type` lock substring (must remain unlocked and fail before plan);
+  - one local failure fixture for `implement-only` task-level spec work inside `Task Orchestration`;
+  - one local failure fixture for `implement-only` sync-spec task inside `Task Orchestration`;
+  - one local failure fixture for `implement-only` copy-spec-path task inside `Task Orchestration`;
+  - one local failure fixture for `implement-only` sync-spec task that references an existing spec path (still forbidden);
+  - one local failure fixture for `implement-only` update-existing-spec task (still forbidden);
+  - one local failure fixture for `implement-only` create-migrated-spec-copy task from existing spec path (still forbidden);
+  - one local failure fixture for `implement-only` create-spec task with `for tests` wording bypass attempt;
+  - one local failure fixture for `implement-only` update-spec task in a `*-tests` path (should not be treated as test-reference exemption);
+  - one local failure fixture for `implement-only` forbidden skill binding to `$topic-spec`;
+  - one local failure fixture for `implement-only` forbidden skill binding to `$normal-flow`;
+  - one local failure fixture for missing `开发基线` in `Flow Decision`;
+  - one local failure fixture proving a later `开发基线` mention outside `Flow Decision` must not satisfy the required field;
+  - one local failure fixture for abstract `开发基线` wording such as `目标仓库主干`;
+  - one local failure fixture proving repository file/doc paths such as `docs/specs/.../SPEC.md` are not accepted as concrete git baselines;
+  - one local failure fixture proving context-dependent Git aliases such as `@{-1}` are not accepted as concrete git baselines;
+  - one local failure fixture proving remote symbolic HEAD aliases are rejected by stable syntax, not by the checker repo's configured remotes;
+  - one local failure fixture proving placeholder values such as `TBD` are not accepted as concrete git baselines;
+  - one local failure fixture proving a plan must preserve the owner's explicit baseline instead of replacing it with another ref;
+  - one local failure fixture proving a later standalone flow answer (for example `normal` after flow clarification) must not clear an earlier owner-explicit baseline;
+  - one local failure fixture proving a later standalone flow restatement in an already locked planning session must not clear an earlier owner-explicit baseline;
+  - one local failure fixture proving plain Chinese owner wording such as `开发基线使用 upstream/release-2026q3` is enforced as an explicit baseline when the plan replaces it;
+  - one local failure fixture proving plain Chinese owner wording such as `开发基线用 upstream/release-2026q3` is also enforced as an explicit baseline when the plan replaces it;
+  - one local failure fixture proving directional owner wording such as ``基于 \``upstream/release-2026q3\` 开发` is also enforced as an explicit baseline when the plan replaces it;
+  - one local failure fixture proving `BASE_REF=...` owner wording is enforced as an explicit baseline when the plan replaces it;
+  - one local failure fixture proving English owner wording such as `base branch is ...` is also enforced as an explicit baseline when the plan replaces it;
+  - one local failure fixture proving legal trailing ref punctuation such as `release/v1.0!` is not silently normalized away during baseline comparison;
+  - one local failure fixture proving short plain-text refs such as `qa` are still enforced as explicit baselines when the plan replaces them;
+  - one local failure fixture proving a plain-text baseline that legally ends with `;` is not silently normalized away during baseline comparison;
+  - one local failure fixture proving a scope clarification like `计划页面上传 API` does not clear an earlier explicit baseline just because it contains the word `计划`;
+  - one `fast-track` positive case;
+  - one positive case for profile-bearing read-only sub-agents;
+  - one positive case proving explicit read-only prohibitions are not treated as shared writes;
+  - one positive case proving ordinary Chinese review wording such as `检查当前功能实现` is not treated as a write;
+  - one positive case proving descriptive English wording such as `source code implements the required behavior` is not treated as a write;
+  - one positive case proving `inspect commit history in the repository` remains read-only;
+  - one positive case proving non-review read-only analysis sidecars may still carry explicit profiles;
+  - one positive case proving a Tier 3 runtime strategy records readiness and profile guidance without a reviewer schedule;
+  - failure fixtures proving delegated or `main-agent` formal reviewer/lane/profile predeclarations report `FORMAL_REVIEW_PREDECLARATION_FORBIDDEN`;
+  - a raw-plan failure fixture proving direct plans must declare a valid `Tier 1` through `Tier 4` review tier;
+- one positive case for Initiative `task:<name>` executors with profiles;
+- one positive case proving every Initiative `task:<name>` has exactly one parseable acceptance/evidence/owner-gate mapping;
+- failure fixtures for missing, duplicate, unknown-task, incompatible-gate, and non-actionable owner-gated acceptance mappings;
+  - one failure fixture for missing delegated-agent profiles;
+  - one failure fixture for unsupported model/effort profiles;
+  - one failure fixture for an inline profile on `main-agent`;
+  - one failure fixture for legacy free-text `model=...` markers in new output;
+  - failure fixtures for a profile model or effort that violates the bound `$review-agent` matrix;
+  - one failure fixture for `task:<name>` in a non-Initiative flow;
+  - failure fixtures for Chinese/English shared-workspace writes, generic English repository/file/code/functionality/solution writes, inflected implementation verbs, command-style patch/remove/permission writes, task-local/filesystem writes, and contradictory write instructions;
+  - one failure fixture for the rejected `task-agent:<name>` alias;
+  - one legacy recovery fixture for an old unprofiled `sub-agent:<name>`;
+  - one `normal` positive case;
+  - one positive case for no-signal `default_fast_track` auto-lock;
+  - one positive case for default target-repo mainline planning that still emits a concrete ref such as `origin/main`;
+  - one positive case proving generic `基于 <spec/doc>` wording is not treated as an owner-explicit git baseline;
+  - one positive case preserving an explicit English owner baseline phrased as `base branch is ...`;
+  - one positive case preserving an explicit owner branch baseline even when flow_type is answered in a later standalone reply;
+  - one positive case preserving an explicit owner branch baseline even when the owner later repeats the same flow token in an already locked planning session;
+  - one positive case preserving an explicit owner branch baseline expressed through directional wording such as ``基于 \``...\` 开发`;
+  - one positive case proving directional product wording such as `基于 new plan file upload API 规划这个任务` does not get misclassified as an owner-explicit git baseline;
+  - one positive case preserving an explicit owner branch baseline expressed as `BASE_REF=...`;
+  - one positive case preserving an explicit owner branch baseline;
+  - one positive case preserving an explicit owner branch baseline that legally ends with punctuation such as `!`;
+  - one positive case preserving an explicit owner branch baseline that legally ends with punctuation such as `!` even when the plain-text request continues with a comma-delimited follow-up clause;
+  - one positive case preserving an explicit owner branch baseline expressed as a short plain-text ref such as `qa`;
+  - one positive case preserving an explicit owner branch baseline that legally ends with `;` in plain text;
+  - one positive case preserving an explicit owner branch baseline expressed in plain Chinese text without backticks or a colon;
+  - one positive case preserving an explicit owner branch baseline expressed in plain Chinese text with `开发基线用 ...` wording;
+  - one positive case preserving an explicit owner branch baseline even when the plain-text request continues with a comma-delimited follow-up clause;
+  - one positive case preserving an explicit owner branch baseline that uses legal git-ref special characters such as `+`;
+  - one positive case preserving an explicit owner branch baseline whose final path component is `HEAD`, such as `feature/HEAD`;
+  - one positive case preserving an explicit owner branch baseline whose ref contains `to` as part of the branch name;
+  - one positive case preserving an explicit owner branch baseline when the later `flow_type` answer is given as natural-language prose instead of a bare standalone token;
+  - one positive case preserving an explicit owner branch baseline whose ref exactly equals a placeholder-like token such as `todo`;
+  - one positive case preserving an explicit owner branch baseline whose ref syntactically resembles a doc path, such as `feature/README.md`;
+  - one positive case preserving an explicit owner commit baseline;
+  - one positive case proving a generic “please give me a new executable plan” re-ask resets the previous owner baseline and flow lock instead of leaking them into the next plan;
+  - one positive case proving a new explicit `$dev-plan` request that contains `merge-ready` wording still resets the previous owner baseline instead of misclassifying that wording as plain continuation;
+  - one positive case proving a new plain-language request such as `请帮我规划这个任务：...` resets the previous owner baseline instead of being validated against stale session state;
+- one positive case proving a new explicit `$dev-plan` request for a task name such as `merge queue API` is not misclassified as plain continuation;
+- one positive case proving an explicit new `$dev-plan` request still resets the planning session even when the same sentence also contains continuation wording such as `继续`;
+- one positive case proving a new explicit `$dev-plan` request resets the previous request's owner baseline instead of leaking it into a later independent plan;
+  - one positive case proving a new independent plan request can reset the previous owner baseline even when the later user turn does not repeat the `$dev-plan` token;
+  - one positive case proving an equivalent remote ref expansion such as `refs/remotes/origin/main` does not trip owner baseline mismatch when the owner asked for `origin/main`;
+  - one positive case proving ordinary scope clarification wording such as `请按 fast-track 把 plan.md cleanup 也规划进去` does not reset an existing planning session's flow lock or explicit baseline;
+  - one positive case proving ordinary scope clarification wording such as `new plan file upload API` does not reset an existing planning session's flow lock or explicit baseline;
+  - one positive case proving ordinary scope clarification wording such as `请按 normal 把计划页面上传 API 也规划进去，flow 和开发基线都不变` does not reset an existing planning session's flow lock or explicit baseline;
+  - one positive case proving ordinary scope clarification wording such as `请把导入页面也规划进去，其他不变` does not reset an existing planning session's flow lock or explicit baseline;
+  - one positive case proving indirect read-only prohibitions such as `not allowed to`, `forbidden from`, and `不允许` are accepted;
+  - one positive case where a direct non-default flow-skill reference overrides the default fast-track path;
+  - one positive case where a standalone flow token still locks correctly even when followed by normal punctuation;
+  - one positive case with `request_user_input` + `flow_type` lock evidence;
+  - one positive case where user explicit lock changes later and latest lock value should win;
+  - one positive case for keyed `flow_type: fast-flow` explicit lock alias;
+  - one positive case proving default Initiative eligibility may be established by the proposed plan's independently acceptable/mergeable Ticket scope and Wave mechanism without requiring owner contract keywords;
+  - one positive case for full-width skill delimiter parsing (e.g. `，`) with valid `$`-prefixed refs;
+  - one positive case for pure `implement-only` orchestration (no spec creation/migration/sync task);
+  - one positive case for `implement-only` test-planning task that references an existing spec path;
+  - one positive case for `implement-only` parser implementation task that references an existing spec path;
+  - one positive case for `implement-only` parser update task that references an existing spec path;
+  - one ambiguous `继续` fallback case (now expected to fail under strict block, with text fallback still listing all four flows).
+  - one negative case proving Git and shell write commands such as `git add`, `git push`, `git branch`, `git tag`, `git update-ref`, `sed -i`, `tee`, and redirection are rejected for sub-agents;
+
+  - one failure case proving bare git commit and Git global-option variants such as git -C . add and git --git-dir=... push are rejected;
+  - one failure case proving git config mutations and git worktree add/remove are rejected;
+  - one positive case proving git apply --check/stat/numstat/summary remain valid read-only validation commands;
+  - one positive case proving git config queries and git worktree list remain valid read-only inspection commands;
+
+Run:
+
+```bash
+bun run skills/bin/plan_contract_check.ts --cases skills/dev-plan/references/replay-cases/CASES.txt
+bun test skills/dev-plan/scripts/test_migrate_task_orchestration.ts
+```
+
+Interpretation:
+
+- `PASS` means the session/fixture matches the Plan contract.
+- `FINDING ... code=...` indicates a policy violation to fix.
+- `CASE_FAIL` means expected/actual mismatch in replay validation.

package/toolkit/skills/dev-plan/references/replay-cases/abstract-development-baseline-combined-phrase.jsonl

@@ -0,0 +1,2 @@
+{"type":"response_item","payload":{"type":"message","role":"user","content":[{"type":"input_text","text":"请用 $dev-plan 给出 fast-track 计划。"}]}}
+{"type":"response_item","payload":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"<proposed_plan>\n# Demo Combined Abstract Development Baseline\n\n## Summary\n- 结论：示例故意把开发基线写成组合后的抽象主干词组。\n\n## Flow Decision\n- `flow_type`: `fast-track`\n- `flow_source`: `default_fast_track`\n- `lock_status`: `locked`\n- 开发基线：`目标仓库主干基线`\n\n## Goals\n- 目标 1: 演示 checker 对组合抽象开发基线的拦截\n  当前计划故意把 `目标仓库主干基线` 填进 `开发基线`，用于验证 checker 不会把它误当成 concrete ref。\n\n## Non-goals\n- 不覆盖显式 owner baseline 的保留规则。\n\n## Task Orchestration\n- wave: 1\n  - main-agent => 规划实现与 merge-ready 收口 (skill: $review-agent + $fast-flow)\n\n## Acceptance & Validation\n- checker 必须拒绝把组合后的抽象主干词组当成最终 `开发基线`。\n\n## Assumptions/Defaults\n- 这是一个抽象值反例：已确定\n</proposed_plan>"}]}}

package/toolkit/skills/dev-plan/references/replay-cases/abstract-development-baseline.jsonl

@@ -0,0 +1,2 @@
+{"type":"response_item","payload":{"type":"message","role":"user","content":[{"type":"input_text","text":"请用 $dev-plan 给出 fast-track 计划。"}]}}
+{"type":"response_item","payload":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"<proposed_plan>\n# Demo Abstract Development Baseline\n\n## Summary\n- 结论：示例故意把开发基线写成抽象主干词。\n\n## Flow Decision\n- `flow_type`: `fast-track`\n- `flow_source`: `default_fast_track`\n- `lock_status`: `locked`\n- 开发基线：`目标仓库主干`\n\n## Goals\n- 目标 1: 演示 checker 对抽象开发基线的拦截\n  当前计划故意只写抽象默认策略词，而不落成具体 ref。\n\n## Non-goals\n- 不覆盖显式 owner baseline 的保留规则。\n\n## Task Orchestration\n- wave: 1\n  - main-agent => 规划实现与 merge-ready 收口 (skill: $review-agent + $fast-flow)\n\n## Acceptance & Validation\n- checker 必须拒绝把抽象主干词当成最终 `开发基线` 的计划输出。\n\n## Assumptions/Defaults\n- 这是一个抽象值反例：已确定\n</proposed_plan>"}]}}

package/toolkit/skills/dev-plan/references/replay-cases/acceptance-follow-up-items-only.jsonl

@@ -0,0 +1,2 @@
+{"type": "response_item", "payload": {"type": "message", "role": "user", "content": [{"type": "input_text", "text": "按 normal 流程计划"}]}}
+{"type": "response_item", "payload": {"type": "message", "role": "assistant", "content": [{"type": "output_text", "text": "<proposed_plan>\n# Acceptance Follow-up Items Only\n\n## Summary\n- 结论：用于验证仅有 Follow-up items 的 Acceptance & Validation 不能通过。\n\n## Flow Decision\n- `flow_type`: `normal`\n- `flow_source`: `user_explicit`\n- `lock_status`: `locked`\n\n## Goals\n- Goal 1: keep acceptance actionable\n  Acceptance must contain a concrete validation line instead of a generic follow-up label.\n\n## Non-goals\n- Do not allow follow-up-only acceptance text.\n\n## Task Orchestration\n- wave: 1\n  - main-agent => implement the demo workflow change (skill: $normal-flow)\n\n## Acceptance & Validation\n- Follow-up items\n\n## Assumptions/Defaults\n- This fixture intentionally exercises follow-up-only acceptance wording: 已确定\n</proposed_plan>"}]}}

package/toolkit/skills/dev-plan/references/replay-cases/acceptance-notes-only.jsonl

@@ -0,0 +1,2 @@
+{"type": "response_item", "payload": {"type": "message", "role": "user", "content": [{"type": "input_text", "text": "按 normal 流程计划"}]}}
+{"type": "response_item", "payload": {"type": "message", "role": "assistant", "content": [{"type": "output_text", "text": "<proposed_plan>\n# Acceptance Notes Only\n\n## Summary\n- 结论：用于验证仅有 Notes 的 Acceptance & Validation 不能通过。\n\n## Flow Decision\n- `flow_type`: `normal`\n- `flow_source`: `user_explicit`\n- `lock_status`: `locked`\n\n## Goals\n- Goal 1: keep acceptance actionable\n  Acceptance must contain a concrete validation line instead of a generic notes label.\n\n## Non-goals\n- Do not allow label-only acceptance text.\n\n## Task Orchestration\n- wave: 1\n  - main-agent => implement the demo workflow change (skill: $normal-flow)\n\n## Acceptance & Validation\n- Notes\n\n## Assumptions/Defaults\n- This fixture intentionally exercises label-only acceptance wording: 已确定\n</proposed_plan>"}]}}

package/toolkit/skills/dev-plan/references/replay-cases/acceptance-tba-only.jsonl

@@ -0,0 +1,2 @@
+{"type":"response_item","payload":{"type":"message","role":"user","content":[{"type":"input_text","text":"按普通流程计划"}]}}
+{"type":"response_item","payload":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"<proposed_plan>\n# Acceptance TBA Only\n\n## Summary\n- 结论：用于验证 Acceptance 里的 TBA 占位会被拦截。\n\n## Flow Decision\n- `flow_type`: `normal`\n- `flow_source`: `user_explicit`\n- `lock_status`: `locked`\n\n## Goals\n- Goal 1: keep validation concrete\n  Require at least one real validation statement before the plan can pass.\n\n## Non-goals\n- Do not leave acceptance unresolved.\n\n## Task Orchestration\n- wave: 1\n  - main-agent => implement the scoped change (skill: $normal-flow)\n\n## Acceptance & Validation\n- TBA\n\n## Assumptions/Defaults\n- Validation method is still being decided: 待确认\n</proposed_plan>"}]}}