@fitlab-ai/agent-infra 0.9.11 → 0.9.12

package/templates/.agents/skills/review-code/reference/output-templates.zh-CN.md

@@ -20,9 +20,27 @@
 - 所选场景必须通过统一 helper 生成 `{next-step-commands}`
 - 计数行固定显示 5 个数字。manual-validation（`{e}`）不影响分支；`人工裁决`（`{h}`）是本阶段 `needs-human-decision` 行数，属于未闭环账本状态，因此 `{h} > 0` 时 `canAdvance=false`，必须按 `.agents/rules/next-step-output.md` 的「人工裁决待办前置块」展开详情，并只输出修订与复审路径。
 
+### 场景 R：最终化停止但结果可见
+
+当 finalizer 失败，或模型因安全门、无进展、重复诊断或紧急熔断停止时，使用本场景，不调用统一 helper，也不输出跨阶段命令。
+
+```text
+任务 {task-id} 审查结果已生成，但生命周期未推进。
+- 审查产物：.agents/workspace/active/{task-id}/{review-artifact}
+- 最后有效 summary/findings：{last-readable-review-result}
+- 本地修复次数：{repairAttempts}
+- 最后诊断：{last-structured-diagnostic}
+- 停止原因：{stop-reason}
+- 完成事件：未发布 | 跨阶段命令：未生成
+
+说明：本次仅停止生命周期推进，已有审查结果仍可查看。请先进行人工处理，或重新运行当前审查技能。
+```
+
+如果 summary 无法安全解析，`{last-readable-review-result}` 必须改为“摘要不可安全解析”，并保留 artifact 路径和原始结构化诊断；不得推算计数或补写结论。
+
 ### 场景 A：通过且无问题
 
-通过后不得按轮次路由。先比较审查快照树 `T` 与基线 `R` 的树，再读取任务的 `prFlow` / `pr_number`；存在 PR 时调用 `agent-infra-internal platform-checks inspect {task-id}`。只选择以下一个互斥出口：
+通过后不得按轮次路由。先比较审查快照树 `T` 与基线 `R` 的树，再读取任务的 `prFlow` / verified `pr_delivery_fact`；存在 PR 时调用 `agent-infra-internal platform-checks inspect {task-id}`。只选择以下一个互斥出口：
 
 - `T != R^{tree}`：场景 A1（提交）。
 - `T == R^{tree}` 且无 PR：`prFlow=disabled` 用场景 A4（完成），否则用场景 A2（创建 PR）。

package/templates/.agents/skills/review-code/reference/report-template.en.md

@@ -21,8 +21,10 @@ Use this template when writing `review-code.md` or `review-code-r{N}.md`.
 - **Reviewer**: {reviewer-name}
 - **Review Time**: {timestamp}
 - **Scope**: {file-count and major modules}
-- **Review Baseline Commit**: {raw reviewed commit R captured once for this round; must equal this round's HEAD}
-- **Reviewed Diff Base**: {raw D used for the complete diff/fingerprint; the merge-base of R and the PR base SHA when a PR is bound, otherwise R}
+- **Review Target Commit**: {target branch SHA M read once from the task-bound remote/base at review start; never overwritten by a later live target}
+- **Reviewed Head**: {local HEAD R captured once for this round; must equal this round's HEAD}
+- **Review Baseline Commit**: {compatibility display of R; must equal Reviewed Head}
+- **Reviewed Diff Base**: {D used for the complete diff/fingerprint; must equal merge-base(R, saved M)}
 - **Reviewed Diff Fingerprint**: {fingerprint field from git-workflow snapshot}
 - **Reviewed Snapshot Tree**: {tree field from git-workflow snapshot}
 - **Overall Verdict**: {Approved / Changes Requested / Rejected} (pick exactly one; combined phrases will fail the verify gate)

package/templates/.agents/skills/review-code/reference/report-template.zh-CN.md

@@ -21,8 +21,10 @@
 - **审查者**：{reviewer-name}
 - **审查时间**：{timestamp}
 - **审查范围**：{file-count and major modules}
-- **审查基线提交**：{本轮一次性捕获的审查提交 R 原文；必须等于本轮 HEAD}
-- **审查差异基线**：{用于完整 diff/fingerprint 的 D 原文；已绑定 PR 时为 R 与 PR base SHA 的 merge-base，否则为 R}
+- **审查目标提交**：{本轮一次性从任务绑定 remote/base 读取的目标分支 SHA M；不可被后续实时目标覆盖}
+- **审查已检视提交**：{本轮一次性捕获的本地 HEAD R；必须等于本轮 HEAD}
+- **审查基线提交**：{R 的兼容显示字段；必须与审查已检视提交相同}
+- **审查差异基线**：{用于完整 diff/fingerprint 的 D；必须等于 merge-base(R, saved M)}
 - **审查差异指纹**：{git-workflow snapshot 输出的 fingerprint 字段}
 - **审查快照树**：{git-workflow snapshot 输出的 tree 字段}
 - **总体结论**：{通过 / 需要修改 / 拒绝}（恰取一个；禁止写组合短语，否则 verify gate 失败）

package/templates/.agents/skills/review-plan/SKILL.en.md

@@ -16,6 +16,7 @@ Review the latest plan artifact and produce `review-plan.md` or `review-plan-r{N
 ## Boundary / Critical Rules
 
 - This skill only reviews plan artifacts and writes a report; it does not modify product code.
+- Before generating task or lifecycle Markdown that will be synchronized to an Issue, read `.agents/rules/sync-content-generation.md` and apply its producer-side constraints; the sync path does not parse or rewrite the body
 - After executing this skill, you **must** immediately update task.md.
 
 Version stamp rule: when creating or updating `task.md` frontmatter, read `.agents/rules/version-stamp.md` first and write or refresh `agent_infra_version`.
@@ -69,7 +70,7 @@ Create `.agents/workspace/active/{task-id}/{review-artifact}`.
 
 ### 6. Update Task Status
 
-After the report, submit each new finding with `agent-infra-internal task-ledger {task-id} finding-upsert --stage plan --review-artifact {review-artifact} --ordinal {n} --severity {blocker|major|minor} --evidence {review-artifact}#{anchor}`; submit prior-response dispositions with `finding-review --id {ledger-id} --status {confirmed|closed|open|needs-human-decision} --evidence {evidence}`. Do not scan ids or edit ledger rows. After all ledger writes, call `agent-infra-internal task-review {task-id} finalize-summary --stage plan --artifact {review-artifact} {execution-flag}` exactly once.
+After the report, submit each new finding with `agent-infra-internal task-ledger {task-id} finding-upsert --stage plan --review-artifact {review-artifact} --ordinal {n} --severity {blocker|major|minor} --evidence {review-artifact}#{anchor}`; submit prior-response dispositions with `finding-review --id {ledger-id} --status {confirmed|closed|open|needs-human-decision} --evidence {evidence}`. Do not scan ids or edit ledger rows. After all ledger writes, make the initial call to `agent-infra-internal task-review {task-id} finalize-summary --stage plan --artifact {review-artifact} {execution-flag}`; after a failure, follow `.agents/rules/local-artifact-repair.md` to decide whether to edit and rerun, and never treat an error type or `changed=false` as automatic authorization.
 
 Bind and reuse this structured mapping from that one response:
 
@@ -79,13 +80,13 @@ Bind and reuse this structured mapping from that one response:
 {unresolved-minor} = stageStatus.unresolvedFindingCounts.minor
 ```
 
-The intent atomically finalizes the report summary and returns the same ledger snapshot. Do not call `stage-status`, replace placeholders manually, or rescan the finding list. If finalization fails or a response field is missing, stop before the completion event. Derive the verdict and next-step branch from the same response's `stageStatus.canAdvance`: only `canAdvance=true` permits `approved`; otherwise use `changes-requested` or `rejected`. Then run `agent-infra-internal task-event {task-id} review-plan.completed --agent {standard-agent-token} --artifact {review-artifact} --verdict {approved|changes-requested|rejected} --blockers {unresolved-blockers} --major {unresolved-major} --minor {unresolved-minor} --manual-validation {n} {execution-flag}`.
+The intent atomically finalizes the report summary and returns the same ledger snapshot. Do not call `stage-status`, replace placeholders manually, or rescan the finding list. After a failure, the model may edit the same controlled artifact and rerun the same intent only when the shared rule's mechanical gates pass; reassess convergence after every failure. The final complete result determines the verdict and counts from that same snapshot: `stageStatus.canAdvance=true` with an Approved conclusion permits cross-stage advancement; `stageStatus.canAdvance=false` still requires `agent-infra-internal task-event {task-id} review-plan.completed --agent {standard-agent-token} --artifact {review-artifact} --verdict {approved|changes-requested|rejected} --blockers {unresolved-blockers} --major {unresolved-major} --minor {unresolved-minor} --manual-validation {n} {execution-flag}`, using `changes-requested` and routing to same-stage revision/review (use `rejected` when the report explicitly rejects). On failure, model stop, lack of progress, or the emergency cap, do not publish a completion event or cross-stage command; use the `repair-stop` scenario in `reference/output-templates.md` to show existing summary/findings, the artifact, actual repair attempts, the last diagnostic, and the stop reason.
 
 `manual-validation` is the data source for the `Manual-validation` count folded into review rows in `ai task log`; do not add a parallel manual-verification field.
 
 If task.md has a valid `issue_number`, run `agent-infra-internal platform-comment sync {task-id} --kind task --agent {standard-agent-token}`, then `agent-infra-internal platform-comment sync {task-id} --kind artifact --artifact {review-artifact} --agent {standard-agent-token}`.
 
-Before writing the summary, the finalization intent checks decision-detail ids; it auto-removes a duplicate only when exactly one formal `[needs-human-decision]` block remains and every other same-id block is clearly short and informal, then reruns the read-only check. Ambiguous duplicates fail closed. If this check or repair fails, stop before the completion event.
+Before writing the summary, the finalization intent checks decision-detail ids; any visible duplicate returns a structured failure and preserves the artifact bytes, while the model decides whether a minimal edit is safe under the shared rule. If the safety gates fail, diagnostics repeat, no byte-level progress occurs, or the emergency cap is reached, stop before the completion event; the stop path must still show the existing review result instead of hiding the artifact.
 
 ### 7. Run Completion Gate
 

package/templates/.agents/skills/review-plan/SKILL.zh-CN.md

@@ -16,6 +16,7 @@ description: >
 ## 行为边界 / 关键规则
 
 - 本技能只审查方案产物并写报告，不修改业务代码
+- 生成会同步到 Issue 的任务或生命周期 Markdown 前，先读取 `.agents/rules/sync-content-generation.md` 并遵循其中的生成端约束；同步端不解析或改写正文
 - 执行本技能后，你**必须**立即更新 task.md
 
 版本戳规则：创建或更新 `task.md` frontmatter 时，先读取 `.agents/rules/version-stamp.md`，并写入或刷新 `agent_infra_version`。
@@ -72,7 +73,7 @@ agent-infra-internal task-snapshot {task-id} --format text
 
 ### 6. 更新任务状态
 
-报告完成后，新 finding 逐条调用 `agent-infra-internal task-ledger {task-id} finding-upsert --stage plan --review-artifact {review-artifact} --ordinal {n} --severity {blocker|major|minor} --evidence {review-artifact}#{anchor}`；复核上一轮响应时调用 `finding-review --id {ledger-id} --status {confirmed|closed|open|needs-human-decision} --evidence {相称证据}`。不得扫描编号或手写账本行。全部账本写入完成后只调用一次 `agent-infra-internal task-review {task-id} finalize-summary --stage plan --artifact {review-artifact} {execution-flag}`。
+报告完成后，新 finding 逐条调用 `agent-infra-internal task-ledger {task-id} finding-upsert --stage plan --review-artifact {review-artifact} --ordinal {n} --severity {blocker|major|minor} --evidence {review-artifact}#{anchor}`；复核上一轮响应时调用 `finding-review --id {ledger-id} --status {confirmed|closed|open|needs-human-decision} --evidence {相称证据}`。不得扫描编号或手写账本行。全部账本写入完成后首次调用 `agent-infra-internal task-review {task-id} finalize-summary --stage plan --artifact {review-artifact} {execution-flag}`；失败后是否继续编辑并重跑，必须遵循 `.agents/rules/local-artifact-repair.md`，不能把失败类型或 `changed=false` 当作自动授权。
 
 从该次返回值绑定并复用以下结构化映射：
 
@@ -82,13 +83,13 @@ agent-infra-internal task-snapshot {task-id} --format text
 {unresolved-minor} = stageStatus.unresolvedFindingCounts.minor
 ```
 
-该 intent 原子最终化报告摘要并返回同一次账本快照；不得再调用 `stage-status`、手工替换占位符或扫描问题清单。最终化失败或返回字段缺失时，停止在完成事件之前。以同一次返回的 `stageStatus.canAdvance` 决定 verdict 和下一步：仅 `canAdvance=true` 可用 `approved`，否则必须用 `changes-requested` 或 `rejected`。随后执行 `agent-infra-internal task-event {task-id} review-plan.completed --agent {standard-agent-token} --artifact {review-artifact} --verdict {approved|changes-requested|rejected} --blockers {unresolved-blockers} --major {unresolved-major} --minor {unresolved-minor} --manual-validation {n} {execution-flag}`。
+该 intent 原子最终化报告摘要并返回同一次账本快照；不得再调用 `stage-status`、手工替换占位符或扫描问题清单。失败后，模型只能在共享规则的机械安全门通过时修改同一个受控 artifact，并完整重跑相同 intent；每次失败都重新判断是否收敛。最终一次完整成功返回决定同一快照的 verdict 和计数：`stageStatus.canAdvance=true` 且结论为 Approved 时允许跨阶段推进；`stageStatus.canAdvance=false` 时仍须执行 `agent-infra-internal task-event {task-id} review-plan.completed --agent {standard-agent-token} --artifact {review-artifact} --verdict {approved|changes-requested|rejected} --blockers {unresolved-blockers} --major {unresolved-major} --minor {unresolved-minor} --manual-validation {n} {execution-flag}`，使用 `changes-requested` 并路由到同阶段修订/复审（报告明确拒绝时使用 `rejected`）。失败、模型停止、无进展或紧急熔断时，不发布完成事件或跨阶段命令，但必须按 `reference/output-templates.md` 的 `repair-stop` 场景展示已有 summary/findings、artifact、实际修复次数、最后诊断和停止原因。
 
 `manual-validation` 是 `ai task log` 中 review 行「人工校验点」（EN `Manual-validation`）计数的数据源；不要新增并行人工验证字段。
 
 如果 task.md 中存在有效的 `issue_number`，调用 `agent-infra-internal platform-comment sync {task-id} --kind task --agent {standard-agent-token}`，再调用 `agent-infra-internal platform-comment sync {task-id} --kind artifact --artifact {review-artifact} --agent {standard-agent-token}`；失败按 `.agents/rules/issue-sync.md` 记录 warning。
 
-最终化 intent 会在写入摘要前检查详情块 ID；只有“一个正式 `[needs-human-decision]` 块 + 明确的简短非正式重复块”才会自动删除重复块并重新做只读检查，无法唯一判断时失败关闭。该检查或自动修复失败时，停止在完成事件之前。
+最终化 intent 会在写入摘要前检查详情块 ID；任何可见重复都返回结构化失败并保持 artifact 字节不变，由模型按共享规则判断是否进行最小编辑。无法通过安全门、诊断重复、没有实际字节变化或达到紧急熔断时，停止在完成事件之前；停止路径仍必须展示已有审查结果，不得吞掉 artifact 内容。
 
 ### 7. 完成校验
 

package/templates/.agents/skills/review-plan/reference/output-templates.en.md

@@ -18,6 +18,24 @@ Rules:
 - The selected scenario must generate `{next-step-commands}` through the shared helper
 - The count line shows 4 numbers. `Human-decision` (`{h}`) counts this stage's `needs-human-decision` rows; because those rows are unresolved, `{h} > 0` means `canAdvance=false`. Expand the "Pending human-decision pre-block" from `.agents/rules/next-step-output.md` and show revision and re-review paths only.
 
+### Scenario R: Finalization stopped with results visible
+
+When the finalizer fails, or the model stops because of a safety gate, lack of progress, repeated diagnostics, or the emergency cap, use this scenario. Do not call the shared helper or output cross-stage commands.
+
+```text
+Task {task-id} review results were generated, but lifecycle advancement stopped.
+- Review artifact: .agents/workspace/active/{task-id}/{review-artifact}
+- Last valid summary/findings: {last-readable-review-result}
+- Local repair attempts: {repairAttempts}
+- Last diagnostic: {last-structured-diagnostic}
+- Stop reason: {stop-reason}
+- Completion event: not published | Cross-stage commands: not generated
+
+Note: only lifecycle advancement stopped; the existing review result remains available. Handle it manually or rerun the current review skill.
+```
+
+If the summary cannot be parsed safely, set `{last-readable-review-result}` to "summary could not be parsed safely" and retain the artifact path and raw structured diagnostic; do not infer counts or add a conclusion.
+
 ### Scenario A: Approved with no findings
 
 Populate `{next-step-commands}` for this scenario by running `agent-infra-internal agent-client next-steps --skill code-task --task-ref {task-ref}`.

package/templates/.agents/skills/review-plan/reference/output-templates.zh-CN.md

@@ -18,6 +18,24 @@
 - 所选场景必须通过统一 helper 生成 `{next-step-commands}`
 - 计数行固定显示 4 个数字。`人工裁决`（`{h}`）是本阶段 `needs-human-decision` 行数；它属于未闭环账本状态，因此 `{h} > 0` 时 `canAdvance=false`，必须按 `.agents/rules/next-step-output.md` 的「人工裁决待办前置块」展开详情，并只输出修订与复审路径。
 
+### 场景 R：最终化停止但结果可见
+
+当 finalizer 失败，或模型因安全门、无进展、重复诊断或紧急熔断停止时，使用本场景，不调用统一 helper，也不输出跨阶段命令。
+
+```text
+任务 {task-id} 审查结果已生成，但生命周期未推进。
+- 审查产物：.agents/workspace/active/{task-id}/{review-artifact}
+- 最后有效 summary/findings：{last-readable-review-result}
+- 本地修复次数：{repairAttempts}
+- 最后诊断：{last-structured-diagnostic}
+- 停止原因：{stop-reason}
+- 完成事件：未发布 | 跨阶段命令：未生成
+
+说明：本次仅停止生命周期推进，已有审查结果仍可查看。请先进行人工处理，或重新运行当前审查技能。
+```
+
+如果 summary 无法安全解析，`{last-readable-review-result}` 必须改为“摘要不可安全解析”，并保留 artifact 路径和原始结构化诊断；不得推算计数或补写结论。
+
 ### 场景 A：通过且无问题
 
 使用 `agent-infra-internal agent-client next-steps --skill code-task --task-ref {task-ref}` 生成本场景的 `{next-step-commands}`。

package/templates/.agents/skills/review-pr/SKILL.en.md

@@ -16,6 +16,7 @@ Based on the target PR's lifecycle-evidence completeness and change risk, select
 - `pr-review*` is an independent artifact family: it does not join the analysis/plan/code stage chain and never changes `current_step`
 - Only a formal PR Review (head SHA, conclusion, findings, receipt, Issue artifact link) is published on the PR; no second full process copy goes to a regular PR comment
 - The full process copy and the reversible `task.md` mirror are synced only to the linked Issue; `restore-task` keeps its Issue-only contract
+- Before generating artifact Markdown that will be synced to an Issue, read `.agents/rules/sync-content-generation.md` and follow its generator-side constraints; Issue sync remains transparent and does not parse or rewrite the body
 - Never backfill or fabricate `analysis`/`plan`/`code` lifecycle history for an external PR
 - The head SHA is recorded only in `pr-review*` and the formal PR Review; on head drift you must start a new round and never reuse an old conclusion
 - Review depth is driven by evidence completeness and change risk, never by the submitter's identity

package/templates/.agents/skills/review-pr/SKILL.zh-CN.md

@@ -16,6 +16,7 @@ description: >
 - `pr-review*` 是独立 artifact family：不接入 analysis/plan/code 阶段依赖链，不修改 `current_step`
 - PR 上只发布正式 PR Review（head SHA、结论、finding、receipt、Issue artifact 链接），不发第二份完整过程原文到 PR 普通评论
 - 完整过程原文与 `task.md` 的可逆镜像只在关联 Issue 上同步；`restore-task` 保持 Issue-only 契约
+- 生成会同步到 Issue 的 artifact Markdown 前，先读取 `.agents/rules/sync-content-generation.md` 并遵循其中的生成端约束；Issue 同步保持透明，不解析或改写正文
 - 不得为外部 PR 倒填或伪造已发生的 `analysis`/`plan`/`code` 生命周期历史
 - head SHA 唯一记录于 `pr-review*` 与正式 PR Review；head 漂移必须进入新轮次，不得沿用旧结论
 - 审查深度由证据完整度和变更风险决定，不由提交者身份决定

package/templates/.agents/skills/review-pr/reference/evidence-grading.en.md

@@ -17,7 +17,7 @@ Evidence scenario classification (S1/S2/S3) → freshness/alignment → risk gra
 
 - Exactly one hit → `unique`; multiple without a unique winner → `ambiguous` (candidate list); zero hits → `none`.
 - Ambiguous hosts fail closed and never enter evidence classification. `pr-review-grade resolve-host --pr <N>` returns a typed `HostResolution`, and `decide` explicitly rejects `ambiguous`.
-- `Closes/Fixes #N` in the PR body (case-insensitive, comma/space separated, lists supported) is parsed by `extractClosingIssueNumbers`; a local `active/*/task.md` `pr_number` hit takes priority over the `issue_number` reverse lookup, and a task hit via both paths deduplicates to a single candidate.
+- `Closes/Fixes #N` in the PR body (case-insensitive, comma/space separated, lists supported) is parsed by `extractClosingIssueNumbers`; a local `active/*/task.md` verified `pr_delivery_fact.identity.number` hit takes priority over the `issue_number` reverse lookup, and a task hit via both paths deduplicates to a single candidate.
 
 ## Evidence Scenario Classification (S1 → S2 → S3)
 
@@ -33,7 +33,7 @@ Evidence scenario classification (S1/S2/S3) → freshness/alignment → risk gra
 ## Freshness and Alignment
 
 - Freshness benchmark = the reviewed head SHA recorded by the latest `pr-review*`, compared character-for-character with the current head: match → `fresh`, otherwise → `stale`.
-- Alignment = freshness is `fresh` and the `issue_number` / `pr_number` match the resolution result.
+- Alignment = freshness is `fresh` and the `issue_number` / verified `pr_delivery_fact.identity.number` match the resolution result.
 - No prior `pr-review*` → `n/a` / `n/a` (first-review special case).
 
 ## Risk Grading (pure evidence)

package/templates/.agents/skills/review-pr/reference/evidence-grading.zh-CN.md

@@ -17,7 +17,7 @@
 
 - 唯一命中 → `unique`；多个且无法唯一 → `ambiguous`（候选列表）；零命中 → `none`。
 - 多宿主歧义即 fail closed：不进入证据分类；由 `pr-review-grade resolve-host --pr <N>` 返回 typed `HostResolution`，`decide` 对 `ambiguous` 显式拒绝。
-- PR body 的 `Closes/Fixes #N`（大小写不敏感、逗号/空格分隔、支持列表）经 `extractClosingIssueNumbers` 解析；本地 `active/*/task.md` 的 `pr_number` 命中优先于 `issue_number` 反查，同一任务经两条路径命中按单候选去重。
+- PR body 的 `Closes/Fixes #N`（大小写不敏感、逗号/空格分隔、支持列表）经 `extractClosingIssueNumbers` 解析；本地 `active/*/task.md` 的 verified `pr_delivery_fact.identity.number` 命中优先于 `issue_number` 反查，同一任务经两条路径命中按单候选去重。
 
 ## 证据场景分类（S1 → S2 → S3）
 
@@ -33,7 +33,7 @@
 ## 新鲜度与对齐
 
 - 新鲜度基准 = 上一轮 `pr-review*` 记录的被审 head SHA，与当前 head 逐字符比对：一致 → `fresh`，否则 → `stale`。
-- 对齐 = freshness 为 fresh 且 `issue_number` / `pr_number` 与解析结果一致。
+- 对齐 = freshness 为 fresh 且 `issue_number` / verified `pr_delivery_fact.identity.number` 与解析结果一致。
 - 无先前 `pr-review*` → `n/a` / `n/a`（首轮特例）。
 
 ## 风险分级（纯证据口径）

package/templates/.agents/skills/run-manual-validation/SKILL.en.md

@@ -18,6 +18,7 @@ The entry point may omit the task ref; explicit task scope accepts only `--task
 - `complete-manual-validation` remains the maintainer's final registration after coverage is judged sufficient.
 - Never manipulate temporary worktrees, leases, or containers directly; only call `agent-infra-internal task-validate`.
 - Never record tokens, environment variables, full argv, absolute user paths, or raw transcripts in the artifact.
+- Before generating a validation artifact Markdown file that will be synced to an Issue, read `.agents/rules/sync-content-generation.md` and follow its generator-side constraints; Issue sync remains transparent and does not parse or rewrite the body.
 
 ## Step 0: State Check (pre-execution hard gate)
 

package/templates/.agents/skills/run-manual-validation/SKILL.zh-CN.md

@@ -18,6 +18,7 @@ description: >
 - `complete-manual-validation` 仍是维护者确认覆盖充分后的最终登记入口。
 - 禁止直接操作临时 worktree、lease 或 container；只调用 `agent-infra-internal task-validate`。
 - 产物不得记录 token、环境变量、完整 argv、绝对用户路径或原始 transcript。
+- 生成会同步到 Issue 的验证 artifact Markdown 前，先读取 `.agents/rules/sync-content-generation.md` 并遵循其中的生成端约束；Issue 同步保持透明，不解析或改写正文。
 
 ## 第 0 步：状态核对（执行前硬约束）
 

package/templates/.agents/skills/run-task/SKILL.en.md

@@ -18,7 +18,7 @@ The orchestrator delegates only and never executes a stage skill itself. First r
 3. Run `task-orchestration {task-id} begin-or-resume --client {client}` with the complete explicit policy. With no explicit policy, core reads that client's configured policy; existing runs use persisted policy. Disk state outside the complete current structure fails closed without rewriting; finish or clear active runs before upgrading. Stop on paused/completed results.
 4. Run `route`. On `completed`, run `agent-infra-internal task-verify {task-id} run-task.completed --format text` and stop. Only a `running` result with non-null `next` supplies action, role, round, artifact, requested model, and requested effort.
 5. Codex runs `codex-lifecycle capability-arm --task-id {task-id}` once. The current loop's real PostToolUse must attest that ordinary tool response. Pass its marker token, the exact routed model, and the exact routed reasoning effort to `prepare --client {client} --requested-model {requestedModel} --requested-reasoning-effort {requestedReasoningEffort} --capability-token {token}`. Core first validates the exact routed model/effort, capability provenance, and controller binding, then captures a read-only snapshot, builds an in-memory prepared receipt, atomically consumes the token, and saves the prepared state with the capability session, turn, and tool-use origin. Activation then verifies that spawn belongs to the same session and turn and uses a distinct tool-use. Other clients use the same prepare without a token. Failure creates no baseline, receipt, or child.
-6. After prepare, run `task-orchestration <task-ref> dispatch` immediately before calling the fresh native child, then use route's exact model/effort and only the short task ref, skill, `--orchestrated`, and stage identity. The trusted hook-spawn first-observation time must fall between dispatch and the deadline. Its first provenance-sensitive command must be `await-activation --stage ... --round ... --artifact ... --role ...`; before `running`, it must not snapshot, emit stage-started, write business files, or start commit. Only explicit `recover-prepared` may clear an expired orphan when the task fingerprint is exact and no matching unconsumed active lifecycle evidence remains.
+6. After prepare, run `task-orchestration <task-ref> dispatch` immediately before calling the fresh native child, then use route's exact model/effort and only the short task ref, skill, `--orchestrated`, and stage identity. The trusted hook-spawn first-observation time must fall between dispatch and the deadline. Its first provenance-sensitive command must be `await-activation --stage ... --round ... --artifact ... --role ...`; before `running`, it must not snapshot, emit stage-started, or write business files. The code stage may create a local checkpoint only through code-task's durable intent core; it must not delegate a separate commit stage or push. Only explicit `recover-prepared` may clear an expired orphan when the task fingerprint is exact and no matching unconsumed active lifecycle evidence remains.
 7. Codex uses SubagentStart/Stop and App Server actual evidence to activate, consume, and seal one receipt; trusted parent fallback must produce the same complete provenance. Timed-out waits do nothing. Advance only a sealed receipt and repeat step 4 only while running.
 8. Create a fresh child each round and never follow up with an old reviewer. Each requested/actual mismatch needs its own host fallback reason. Protocol, capability, receipt hook/evidence binding, source, controller, identity, ledger, or fingerprint failure pauses and fails closed; cross-root package/build/contract or hook/profile content drift is delivered as a warning instead of blocking natural evolution.
 9. On pause or completion, run the matching typed verification and report the structured endpoint.

package/templates/.agents/skills/run-task/SKILL.zh-CN.md

@@ -18,7 +18,7 @@ description: >
 3. 调用 `agent-infra-internal task-orchestration {task-id} begin-or-resume --client {client}` 并转发完整显式策略。完全没有显式策略时由核心读取当前 client 的 `agentClients[].orchestration`；existing run 使用持久化策略。磁盘状态不符合当前完整结构时核心失败关闭且不改写；升级前必须完成或清空 active run。仅当核心返回 `ORCHESTRATION_MODEL_POLICY_REQUIRED` 时，先用 `agent-client model-selection` 展示 complete/partial/interactive-only 来源，再一次收集完整策略；未回答则不创建 run。若为 paused/completed，按结构化结果停止。
 4. 调用 `route` 并读取结构化结果。若返回 `completed`，立即运行 `agent-infra-internal task-verify {task-id} run-task.completed --format text` 并停止；仅当返回 `running` 且 `next` 非空时读取唯一 action、role、round、artifact、`requestedModel` 和 `requestedReasoningEffort`，不得自行推断。
 5. Codex 先调用一次 `agent-infra-internal codex-lifecycle capability-arm --task-id {task-id}`。该普通工具调用必须由当前 loop 的真实 PostToolUse 回写 attestation；把 route 返回的精确 model、effort 和 marker 中 token 一并传给 `prepare --client {client} --requested-model {requestedModel} --requested-reasoning-effort {requestedReasoningEffort} --capability-token {token}`。核心先校验精确的 model/effort、capability provenance 与 controller binding，再只读捕获快照、构造内存中的 prepared receipt，最后原子消费 token 并保存 prepared 状态，同时持久化 capability 的 session、turn、tool-use 来源。activation 再校验 spawn 与该来源属于同一 session/turn，且 spawn 使用独立的 tool-use。其他客户端直接调用同一 prepare（无 token）。任一失败不得创建 baseline、receipt 或 child。
-6. prepare 成功后，在调用 fresh 原生子 Agent 的紧邻前一步执行 `task-orchestration <task-ref> dispatch`，再显式覆盖 route 返回的 model/effort；只传短任务引用、skill 名、`--orchestrated` 与 stage/round/artifact/role。可信 hook-spawn 首次观察时间必须位于 dispatch 与 deadline 之间。child 的第一条 provenance-sensitive 命令必须是 `await-activation --stage ... --round ... --artifact ... --role ...`；返回 running 前禁止 snapshot、阶段 started、业务写入或 commit。超时由核心暂停；崩溃遗留 prepared receipt 只能在 exact workspace fingerprint、无匹配的未消费 active lifecycle evidence 且 deadline 已过时显式 `recover-prepared`。
+6. prepare 成功后，在调用 fresh 原生子 Agent 的紧邻前一步执行 `task-orchestration <task-ref> dispatch`，再显式覆盖 route 返回的 model/effort；只传短任务引用、skill 名、`--orchestrated` 与 stage/round/artifact/role。可信 hook-spawn 首次观察时间必须位于 dispatch 与 deadline 之间。child 的第一条 provenance-sensitive 命令必须是 `await-activation --stage ... --round ... --artifact ... --role ...`；返回 running 前禁止 snapshot、阶段 started 或业务写入；代码阶段的本地 checkpoint 只能通过 code-task 的持久化 intent 核心完成，禁止单独委派 commit 阶段或推送。超时由核心暂停；崩溃遗留 prepared receipt 只能在 exact workspace fingerprint、无匹配的未消费 active lifecycle evidence 且 deadline 已过时显式 `recover-prepared`。
 7. Codex 用 SubagentStart/Stop 与 App Server actual evidence 激活、消费并封存唯一 receipt；受信 parent fallback 仍必须形成相同的完整 provenance。timed-out wait 无动作；只有 empty turns 或协议 `inProgress` 可等待，malformed、身份/传输错误或异常 terminal 均暂停。子 Agent 返回后只对 sealed receipt 调用 `advance`；只有 `running` 才重复步骤 4。
 8. 每轮创建新 child；禁止 follow-up 复用 reviewer。actual model/effort 与 requested 不同时必须有各自的宿主降级理由。protocol、capability、receipt 内 hook/evidence binding、source、controller、身份、账本或 fingerprint 异常都调用 `pause` 并失败关闭；跨根 package/build/contract 或 hook/profile 内容漂移按 warning 交付，不把自然演进差异当成硬失败。
 9. 完成或暂停后运行对应 typed verification，并把结构化 run 摘要、暂停原因、commit 终点或 clean completion evidence 告知用户。

package/templates/.agents/skills/update-agent-infra/scripts/sync-templates.js

@@ -74,6 +74,10 @@ const DEFAULTS = {
   "task": {
     "shortIdLength": 2
   },
+  "delivery": {
+    "remote": "origin",
+    "baseRef": "main"
+  },
   "labels": {
     "in": {}
   },

package/templates/.agents/skills/watch-pr/SKILL.en.md

@@ -39,8 +39,8 @@ After prerequisites pass and before this round's first artifact action, append a
 
 Resolve the target PR number `{pr#}` and an optional `{task-id}` via these deterministic branches:
 
-- Scenario A (argument omitted): reverse-lookup the active task from the current branch, then read its `pr_number`.
-- Scenario B (omitted task ref or `--task/-t`, **task-anchored primary path**): resolve the full `{task-id}` via "Task Context Resolution" and read `.agents/workspace/active/{task-id}/task.md` for `pr_number` as `{pr#}`; if `pr_number` is empty, follow "Error Handling" to prompt running `create-pr` first, then stop.
+- Scenario A (argument omitted): reverse-lookup the active task from the current branch, then read its verified `pr_delivery_fact.identity.number`.
+- Scenario B (omitted task ref or `--task/-t`, **task-anchored primary path**): resolve the full `{task-id}` via "Task Context Resolution" and read `.agents/workspace/active/{task-id}/task.md` for verified `pr_delivery_fact.identity.number` as `{pr#}`; if the fact is unbound, follow "Error Handling" to prompt running `create-pr` first, then stop.
 - Scenario C (`--pr <number>` or a PR URL): use that PR number directly as `{pr#}`; then determine `{task-id}` via "Reverse-lookup task".
 - Reverse-lookup task (scenarios A / C): use task context/query capabilities to find the unique active task bound to `{pr#}`. If none exists, stop and request binding first; the typed checks intent does not create a second taskless state machine.
 
@@ -74,7 +74,7 @@ Update `.agents/workspace/active/{task-id}/task.md`:
 - `assigned_to`: {current agent}
 - `updated_at`: {current time}
 - `agent_infra_version`: per `.agents/rules/version-stamp.md`
-- **Do not change** `pr_status` (keep `created`) or `current_step`
+- **Do not change** `pr_delivery_fact` or `current_step`
 - **Append** to `## Activity Log` (do not overwrite prior entries; `{N}` = number of existing Watch PR entries for this task + 1):
   ```
   - {YYYY-MM-DD HH:mm:ss±HH:MM} — **Watch PR (Round {N})** by {agent} — {success: PR ready, repair commits: {k} [{SHA summary}] / blocked: blocked: {summary}}
@@ -148,6 +148,6 @@ Stop immediately after the checklist. The green exit waits for the user to run `
 
 ## Error Handling
 
-- Cannot locate a PR (task short id resolves but task.md has no `pr_number`, and no `--pr` was passed and the current branch has no PR): prompt "Run `create-pr` first, or specify the PR with `--pr <number>`", then stop.
+- Cannot locate a PR (task short id resolves but task.md has no verified bound `pr_delivery_fact`, and no `--pr` was passed and the current branch has no PR): prompt "Run `create-pr` first, or specify the PR with `--pr <number>`", then stop.
 - Platform CLI not authenticated or API unavailable: prompt that manual intervention is needed, then stop.
 - Short-id resolution failure: pass through `task-short-id.js`'s exit code and error message; do not rewrite it.

package/templates/.agents/skills/watch-pr/SKILL.zh-CN.md

@@ -39,8 +39,8 @@ description: >
 
 按以下确定性分支解析出目标 PR 号 `{pr#}` 与可选 `{task-id}`：
 
-- 场景 A（省略入参）：从当前分支反查 active task；定位后读取其 `pr_number`。
-- 场景 B（省略 task ref 或 `--task/-t`，**任务锚定主路径**）：按「任务上下文解析」取得完整 `{task-id}`。读 `.agents/workspace/active/{task-id}/task.md` 取 `pr_number` 作为 `{pr#}`；`pr_number` 为空时按「错误处理」提示先 `create-pr`，停止。
+- 场景 A（省略入参）：从当前分支反查 active task；定位后读取其 verified `pr_delivery_fact.identity.number`。
+- 场景 B（省略 task ref 或 `--task/-t`，**任务锚定主路径**）：按「任务上下文解析」取得完整 `{task-id}`。读 `.agents/workspace/active/{task-id}/task.md` 取 verified `pr_delivery_fact.identity.number` 作为 `{pr#}`；fact 未绑定时按「错误处理」提示先 `create-pr`，停止。
 - 场景 C（`--pr <number>` 或 PR URL）：直接取该 PR 号为 `{pr#}`；随后按「反查任务」确定 `{task-id}`。
 - 反查任务（场景 A / C）：通过任务上下文/任务查询取得与 `{pr#}` 唯一绑定的 active task。未命中时停止并提示先绑定 PR；typed checks intent 不建立第二套无任务状态机。
 
@@ -74,7 +74,7 @@ date "+%Y-%m-%d %H:%M:%S%z" | sed 's/\([+-][0-9][0-9]\)\([0-9][0-9]\)$/\1:\2/'
 - `assigned_to`：{当前代理}
 - `updated_at`：{当前时间}
 - `agent_infra_version`：按 `.agents/rules/version-stamp.md` 取值
-- **不改** `pr_status`（保持 `created`）与 `current_step`
+- **不改** `pr_delivery_fact` 与 `current_step`
 - **追加**到 `## 活动日志`（不要覆盖之前的记录；`{N}` = 本任务已有 Watch PR 条目数 + 1）：
   ```
   - {YYYY-MM-DD HH:mm:ss±HH:MM} — **Watch PR (Round {N})** by {agent} — {成功：PR ready, repair commits: {k} [{sha 摘要}] / 阻塞：blocked: {简述}}
@@ -148,6 +148,6 @@ agent-infra-internal task-verify {task-id} watch-pr.completed --format text
 
 ## 错误处理
 
-- 无法定位 PR（任务短号命中但 task.md 无 `pr_number`，且未传 `--pr`、当前分支也无 PR）：提示「请先运行 `create-pr`，或用 `--pr <number>` 指定 PR」，停止。
+- 无法定位 PR（任务短号命中但 task.md 无 verified bound `pr_delivery_fact`，且未传 `--pr`、当前分支也无 PR）：提示「请先运行 `create-pr`，或用 `--pr <number>` 指定 PR」，停止。
 - 平台 CLI 未认证或 API 不可用：提示需人工介入，停止。
 - 短号解析失败：透传 `task-short-id.js` 的退出码与错误信息，不重写。

package/templates/.agents/templates/task.en.md

@@ -13,7 +13,11 @@ start_date:                     # Optional Issue field for Feature: YYYY-MM-DD
 target_date:                    # Optional Issue field for Feature: YYYY-MM-DD
 current_step: requirement-analysis # requirement-analysis | requirement-analysis-review | technical-design | technical-design-review | code | code-review | completed
 assigned_to:                   # claude | codex | antigravity | opencode | human
-pr_status: pending             # PR status: pending (default) | created (PR created) | skipped (explicitly skipped)
+pr_delivery_fact: '{"version":1,"state":"unbound","reason":"initial"}'
+delivery_remote: origin        # Git remote used to deliver the task branch
+delivery_base_ref: main        # Target branch for the task PR
+checkpoint_commit:             # Most recent local checkpoint commit
+delivery_remote_head:          # Most recent successfully delivered task branch SHA
 ---
 
 # Task: [Title]

package/templates/.agents/templates/task.zh-CN.md

@@ -13,7 +13,11 @@ start_date:                     # Feature 可选 Issue 字段：YYYY-MM-DD
 target_date:                    # Feature 可选 Issue 字段：YYYY-MM-DD
 current_step: requirement-analysis # requirement-analysis | requirement-analysis-review | technical-design | technical-design-review | code | code-review | completed
 assigned_to:                   # claude | codex | antigravity | opencode | human
-pr_status: pending             # PR 状态：pending（默认）| created（已创建 PR）| skipped（显式跳过）
+pr_delivery_fact: '{"version":1,"state":"unbound","reason":"initial"}'
+delivery_remote: origin        # 任务分支交付使用的 Git remote
+delivery_base_ref: main        # 任务 PR 的目标分支
+checkpoint_commit:             # 最近一次本地 checkpoint commit
+delivery_remote_head:          # 最近一次成功交付到 remote 的任务分支 SHA
 ---
 
 # 任务：[标题]

package/templates/.agents/workflows/bug-fix.en.yaml

@@ -3,7 +3,7 @@
 #
 # Note: whether a step's `pr_tasks` list counts toward workflow progress follows the "PR path" rule:
 # `.agents/.airc.json:prFlow="required"` always counts; `prFlow="disabled"` never counts;
-# when the field is absent, exclude only if task.md's `pr_status=skipped`, otherwise count
+# when the field is absent, exclude only if task.md's `pr_delivery_fact.state=skipped`, otherwise count
 # (see .agents/skills/complete-task/SKILL.md).
 
 name: bug-fix
@@ -126,6 +126,7 @@ steps:
       - Add a regression test that fails without the fix
       - Fix findings from the latest code-review artifact when in fix mode
       - Verify existing tests still pass
+      - Create a local checkpoint commit after tests pass
       - Ensure code follows project conventions
     inputs:
       - Approved fix plan
@@ -177,21 +178,21 @@ steps:
       - Code review report
       - Findings list, if any
 
-  - name: commit
-    description: Finalize the defect fix and create a pull request (the PR portion runs only on the PR path).
+  - name: delivery
+    description: Deliver the approved checkpoint as a pull request, then complete the task after merge and final gates.
     recommended_agents:
       - claude
       - human
     tasks:
-      - Ensure all tests pass
-      - Write a commit message that references the defect or Issue
-      - Move the task to completed
+      - Verify the current HEAD still equals the approved checkpoint
+      - Confirm the task-bound delivery remote and base branch remain valid
+      - Move the task to completed after PR merge and final gates
     pr_tasks:
       - Create a pull request with a defect-fix description
       - Link the PR to the defect report Issue
     inputs:
-      - Final bug-fix branch
+      - Approved local checkpoint
       - Task workspace
     outputs:
-      - Completed task workspace under .agents/workspace/completed/
       - Pull request (only on the PR path)
+      - Completed task workspace under .agents/workspace/completed/ after merge and final gates

package/templates/.agents/workflows/bug-fix.zh-CN.yaml

@@ -3,7 +3,7 @@
 #
 # 注：步骤中的 `pr_tasks` 列表按「走 PR 路径」判定是否计入工作流进度：
 # `.agents/.airc.json:prFlow="required"` 始终计入；`prFlow="disabled"` 不计入；
-# 字段缺省时仅当 task.md 的 `pr_status=skipped` 排除，否则计入（详见 .agents/skills/complete-task/SKILL.md）。
+# 字段缺省时仅当 task.md 的 `pr_delivery_fact.state=skipped` 排除，否则计入（详见 .agents/skills/complete-task/SKILL.md）。
 
 name: bug-fix
 description: 诊断和修复缺陷的工作流。
@@ -125,6 +125,7 @@ steps:
       - 添加在没有修复时会失败的回归测试
       - 修复最新 code-review 产物中要求处理的问题（如处于修复模式）
       - 验证现有测试仍然通过
+      - 测试通过后创建本地 checkpoint commit
       - 确保代码遵循项目规范
     inputs:
       - 已通过审查的修复方案
@@ -176,21 +177,21 @@ steps:
       - 代码审查报告
       - 问题列表（如有）
 
-  - name: commit
-    description: 最终确认缺陷修复并创建拉取请求（PR 部分仅在走 PR 路径时执行）。
+  - name: delivery
+    description: 将已审查的 checkpoint 交付为 PR，并在合并及最终门禁通过后完成任务。
     recommended_agents:
       - claude
       - human
     tasks:
-      - 确保所有测试通过
-      - 编写引用缺陷/Issue 的提交信息
-      - 将任务移至已完成
+      - 验证当前 HEAD 仍等于已审查的 checkpoint
+      - 确认任务绑定的交付 remote 和目标分支仍有效
+      - 在 PR 合并及最终门禁通过后将任务移至已完成
     pr_tasks:
       - 创建包含缺陷修复描述的拉取请求
       - 将 PR 关联到缺陷报告 Issue
     inputs:
-      - 最终缺陷修复分支
+      - 已审查的本地 checkpoint
       - 任务文件
     outputs:
-      - 已完成的任务文件（位于 .agents/workspace/completed/）
       - 拉取请求（仅走 PR 路径时）
+      - 已完成的任务文件（PR 合并和最终门禁通过后）

package/templates/.agents/workflows/feature-development.en.yaml

@@ -3,7 +3,7 @@
 #
 # Note: whether a step's `pr_tasks` list counts toward workflow progress follows the "PR path" rule:
 # `.agents/.airc.json:prFlow="required"` always counts; `prFlow="disabled"` never counts;
-# when the field is absent, exclude only if task.md's `pr_status=skipped`, otherwise count
+# when the field is absent, exclude only if task.md's `pr_delivery_fact.state=skipped`, otherwise count
 # (see .agents/skills/complete-task/SKILL.md).
 
 name: feature-development
@@ -126,6 +126,7 @@ steps:
       - Write unit tests
       - Write integration tests when applicable
       - Fix findings from the latest code-review artifact when in fix mode
+      - Create a local checkpoint commit after tests pass
       - Ensure code follows project conventions
     inputs:
       - Approved technical design
@@ -177,21 +178,21 @@ steps:
       - Code review report
       - Findings to fix
 
-  - name: commit
-    description: Finalize changes and create a pull request (the PR portion runs only on the PR path).
+  - name: delivery
+    description: Deliver the approved checkpoint as a pull request, then complete the task after merge and final gates.
     recommended_agents:
       - claude
       - human
     tasks:
-      - Ensure all tests pass
-      - Write a clear commit message following conventions
-      - Move the task to completed
+      - Verify the current HEAD still equals the approved checkpoint
+      - Confirm the task-bound delivery remote and base branch remain valid
+      - Move the task to completed after PR merge and final gates
     pr_tasks:
       - Create a pull request with an appropriate description
       - Link the PR to the related Issue
     inputs:
-      - Final implementation branch
+      - Approved local checkpoint
       - Task workspace
     outputs:
-      - Completed task workspace under .agents/workspace/completed/
       - Pull request (only on the PR path)
+      - Completed task workspace under .agents/workspace/completed/ after merge and final gates

package/templates/.agents/workflows/feature-development.zh-CN.yaml

@@ -3,7 +3,7 @@
 #
 # 注：步骤中的 `pr_tasks` 列表按「走 PR 路径」判定是否计入工作流进度：
 # `.agents/.airc.json:prFlow="required"` 始终计入；`prFlow="disabled"` 不计入；
-# 字段缺省时仅当 task.md 的 `pr_status=skipped` 排除，否则计入（详见 .agents/skills/complete-task/SKILL.md）。
+# 字段缺省时仅当 task.md 的 `pr_delivery_fact.state=skipped` 排除，否则计入（详见 .agents/skills/complete-task/SKILL.md）。
 
 name: feature-development
 description: 开发新功能的端到端工作流。
@@ -125,6 +125,7 @@ steps:
       - 编写单元测试
       - 编写集成测试（如适用）
       - 修复最新 code-review 产物中要求处理的问题（如处于修复模式）
+      - 测试通过后创建本地 checkpoint commit
       - 确保代码遵循项目规范
     inputs:
       - 已通过审查的技术设计
@@ -176,21 +177,21 @@ steps:
       - 代码审查报告
       - 待修复问题列表
 
-  - name: commit
-    description: 最终确认变更并创建拉取请求（PR 部分仅在走 PR 路径时执行）。
+  - name: delivery
+    description: 将已审查的 checkpoint 交付为 PR，并在合并及最终门禁通过后完成任务。
     recommended_agents:
       - claude
       - human
     tasks:
-      - 确保所有测试通过
-      - 按规范编写清晰的提交信息
-      - 将任务移至已完成
+      - 验证当前 HEAD 仍等于已审查的 checkpoint
+      - 确认任务绑定的交付 remote 和目标分支仍有效
+      - 在 PR 合并及最终门禁通过后将任务移至已完成
     pr_tasks:
       - 创建包含适当描述的拉取请求
       - 将 PR 关联到相关 Issue
     inputs:
-      - 最终实现分支
+      - 已审查的本地 checkpoint
       - 任务文件
     outputs:
-      - 已完成的任务文件（位于 .agents/workspace/completed/）
       - 拉取请求（仅走 PR 路径时）
+      - 已完成的任务文件（PR 合并和最终门禁通过后）