跳到正文
cccskills
免费在 GitHub 公开

dough-story-wrap-up

Closes one completed feature story, bounded retrospective correction or context-only planless execution using available execution context. Assimilates lasting product knowledge, settles existing follow-ups and authorized product decisions, and removes spent plan and history with Git recovery. Reports missing or unfinished inputs truthfully. Use to close completed work, delete spent records, or continue settled planned execution under Execute Plan's continuation rule. With `--process-retrospective`, it also reviews the wrap-up's own process and records supported findings in `DearDough.md`.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut星标 50(GitHub 50 + cccskills 0)

实测评分 未实测

查看安装方法

包含的文件(7)

  • SKILL.md15.4 KB
  • references/follow-up-disposition.md3.2 KB
  • references/story-branch-integration.md2.2 KB
  • scripts/trunk-closure-completion.mjs1.0 KB
  • scripts/trunk-closure-observer.mjs6.0 KB
  • scripts/trunk-closure-settlement.mjs6.6 KB
  • scripts/trunk-closure.mjs7.5 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Story wrap-up

Close one selected feature story, bounded retrospective correction, or contextual instruction when the coordinator invokes wrap-up explicitly or through Execute Plan's authorized planned continuation. That continuation retains the same execution context and publication authority, including local-only/no-push instructions. Use available execution context and optional retrospective advice. Leave maintained product knowledge without spent source or plan history in the current snapshot. Do not invent findings, records, or another-conversation requirements. With --process-retrospective, also review this wrap-up's process.

Resolve this project's context

Require one selected work identity. Use a canonical story — a feature story, or a correction's minimal story linked to its plan — when it is that work's active home. A plan-homed correction uses its plan when it satisfies the correction-input contract in planning scope and lifecycle; do not require or create a seed for it. When the source is neither a story nor a correction plan, use a contextual instruction with retained execution identity, changes, and proof; do not require or create a story, plan, or queue entry. Name any missing required field, leave the affected work intact, and stop before closure. Do not invent conventions, artifacts, another review, or claim closure.

Resolve from this project, not this skill's location:

  • repository root and Git working tree;
  • canonical seed location, story identity, heading or stable-anchor conventions for a selected feature story;
  • for planned work: executable-plan location, status vocabulary, and the selected work's plan identity;
  • for planless work: the supplied story or instruction, its changes, available execution results, and retained execution identity;
  • optional retrospective advice when present, including an empty result, and the product advice in the plan's execution-complete record;
  • Git commit conventions used to preserve a recoverable revision;
  • selected mode, execution checkout and branch, authorized remote and target branch, and an originating or default checkout only when one is available;
  • the product backlog path when a Taken or Backlog list entry points at the selected work; and
  • shared records that name the selected work: its seed when applicable, process log (DearDough.md unless this project sets another canonical location), incoming links, and assessment records.

Before deleting a plan that carries planned-execution identity, retain the resolved mode, checkout, branch, remote, target, and the execution checkout's management context in the coordinator's available execution context for the remaining wrap-up actions and report. Do not create a parallel registry. Missing identity needed by a later action stops that action instead of reconstructing or guessing it after plan deletion.

With supplied dashboard reporting context, read and retain dashboard completion, its command and a surviving working directory now, before either Trunk Mode finish or Story Branch retirement can remove the checkout. Prepare any attention file outside it; reporting itself remains the final operation.

Establish execution completion

Judge completion from the selected work and available execution evidence.

  • Planned work is complete when every slice is done.
  • Work without a plan, including an investigation whose approach stayed unselected, is complete when its story or instruction, changes, and execution results show the promised outcome, including evidenced no-change.
  • Incomplete implementation leaves the affected active work intact. Report the unfinished implementation and stop. Do not delete an unfinished correction plan to manufacture a wrap-up.

Assimilate lasting knowledge

Move lasting behavior and design into this project's maintained code, tests, and documentation. Describe the current product without execution narration, impact chronology, story or plan identity, or retrospective judgments. Preserve existing product tests and documents that already state current behavior. Do not invent product knowledge. Before this closure deletes any spent record, preserve its lasting rules.

For North Star topics cited, added, or revised by the completed work, apply the shared topic-retirement instructions as part of ordinary closure.

Settle an existing follow-up plan

When a retrospective follow-up exists, read and apply follow-up disposition before closure: queue it first by default, or drop it only on explicit human instruction. Do not complete wrap-up with an unlisted follow-up.

Apply product-review decisions

When product advice or additional human input is present, apply only authorized compatible backlog and canonical-home changes, following dough-product-backlog queue and active-home conventions. Without retrospective advice in the conversation, as in a fresh session, use the plan's recorded product advice; a recorded skip is a skipped review. Explicit human input wins over either advice.

Supported changes: relevant reorder, queue membership, understood new-story addition, understood bounded-correction addition, and canonical-detail edits. Apply the backlog skill's canonical-home admission rules to every new entry. Keep existing follow-up work first unless a later explicit human instruction changes that priority. Preserve unrelated content, still-needed acceptance work, and near-future direction.

A skipped, empty, or absent product review, or absent extra human input, introduces no mandatory question. Ordinary closure continues using the other available inputs. Leave unresolved necessary context with active work and report the choice; do not invent scope, launch discovery, or start another review.

Commit closure inputs and preserve Git recovery

After supported follow-up queue changes and before deleting anything, make the current revision recoverable with this project's ordinary Git conventions. Commit all uncommitted owned review and closure-input changes in that revision: applicable retrospective edits to the process log in the execution checkout, uncommitted follow-up records (including an explicitly dropped follow-up before its deletion) and queue edits, assimilated product knowledge, and the spent material. Records the execution already committed in its completion commit need no second commit. Preserve unrelated changes and include only files or portions whose ownership is unambiguous. Resolve ownership of the intended cleanup targets at this boundary too, before deleting any of them. Then record that revision as the before-cleanup commit, even when the current revision was already suitable. If commit conventions, ownership, or recovery cannot be resolved, leave the material intact, report the gap, and do not claim closure. In Trunk Mode, publish that commit through wrap-up closure publication before deleting spent history.

Before deletion, follow the shared supplier dependency procedure to discover current consumers and retain recoverable completion evidence. In Trunk Mode, resolve and publish consumer changes after accepted before-cleanup publication and before finish can retire the workspace.

Delete spent history, including shared records

After completion and Git recovery are established — including any required Trunk Mode before-cleanup publication — delete the selected work's spent history under wrap-up cleanup:

  • its executable plan and owned proof, evidence, and assessment records, even when the plan was retained or carries its execution-complete record;
  • its canonical story section when one exists, and its seed only when every remaining section is spent;
  • its Taken or Backlog list entry when one exists, through the dough-product-backlog complete command, which also deletes that work's agent profile, writes its done record, removes expired done records, and rebuilds the done catalog beside them, all owned closure changes; and
  • links whose sole purpose is preserving that history.

Also remove the owned follow-up records selected for explicit drop above, after their Git recovery is established, and their entry with complete --dropped. Otherwise preserve active follow-ups and their queue entries.

Remove empty directories belonging to the spent work, including untracked ones. The current snapshot must be free of that history, both tracked and untracked, with recovery available from the recorded before-cleanup commit.

Preserve process findings, occurrence facts, and unresolved judgments under process-finding recording. Keep findings understandable when removing their spent sources.

Preserve unrelated human text, sibling stories and log entries, product and version identity, maintained tests and documents, still-needed acceptance work, and active follow-ups. Shared records lose only spent portions attributable to the completed work; leave uncertain portions intact and report the ambiguity.

Keep an active follow-up's queue link, correction input, and provenance. If its source locator points into deleted predecessor history, replace the locator with the before-cleanup commit and repository-relative path. Repair Markdown links broken by cleanup without recreating spent history. Repeating wrap-up must recognize already-completed cleanup without duplicating edits. Missing artifacts alone do not establish completion of another work item.

Review this wrap-up's process

Only with --process-retrospective, after spent history is deleted, apply the shared process review of a run once per wrap-up, with the execution checkout as the write location. Point evidence in deleted sources at the before-cleanup commit, as an active follow-up's locator does above. Recorded findings are owned closure changes for Commit final closure and travel with that commit. A rerun that finds the final closure already committed does not review again.

Commit final closure

Review and commit the owned closure changes using this project's Git conventions, preserving unrelated changes. Both the before-cleanup revision and the final closure must be committed in either execution mode, as agent commits when that reference applies; uncommitted cleanup is not completion. Report unresolved ownership or commit failures without claiming closure.

Direct-current-branch mode stays in the recorded checkout and creates no worktree. Local-only closure commits there and reports that revision as committed and pending publication; it does not push. Publish-authorized closure uses wrap-up closure publication from that checkout; the receipt is the accepted SHA and the authorized target. Trunk Mode publishes the final-closure commit through finish; Story Branch Mode continues with integration and resource cleanup below.

Integrate committed Story Branch Mode closure

Read and apply Story Branch integration to the final-closure tip, passing a dashboard landing context supplied to this launch, and require its accepted receipt before resource cleanup.

Remove execution resources safely

Retirement removes the worktree only when its creation record or another record there shows this work created it; pass Dough Land's Retire the worktree ownership facts. In Trunk Mode, finish retires the worktree and branch, never a remote execution branch, once its completion receipt holds success with confirmed shutdown; act on its cleanup result there, and after an interruption rerun it as Finish Trunk Mode closure says. In Story Branch Mode, retire them under that section with --remote-branch <execution branch> and --contained <integrated SHA>, with wrap-up's gate: a success completion receipt for the integrated SHA whose shutdown is confirmed, and no active checkout-bound observer still hosting the worktree under preserve pending local work. Dough Land's refresh never blocks retirement. Report blocked or partial cleanup without repeating already-completed closure. Skip cleanup in direct-current-branch mode.

Report

Apply Dough Land's shared completion attention rule and its final dashboard operation with the supplied launch context, after either closure path has settled. Keep the selected identity, mode and retained execution context, before-cleanup and final-closure commits, accepted publication and completion receipts, remaining coverage, knowledge and cleanup dispositions in their existing operational evidence and lasting homes. Do not produce a success recap solely to repeat them.

When attention is needed, report the affected facts and next action. Distinguish accepted trunk publication from a superseded candidate, an integrated_state success from the closure's own, integration from refused cleanup, committed pending publication from an accepted receipt, and completed wrap-up from a refusal that left files intact. Missing context, unfinished work, unresolved recovery/integration, required push, unpublished Trunk Mode closure, retained or unconfirmed observation, or resource cleanup prevents successful closure. Local-only current-branch pending publication is not a required push.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

backend

免费

Backend Java — Spring Boot controllers/services, JUnit controller tests, MakeMe. Read this first for any backend work. Use when working on the Java Spring backend.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

Backend Java production code, Spring Boot controllers/services/repositories, controller return values, imports, schema ERD pointer. Use when editing backend production Java.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

Backend Java tests, controller-level stable-boundary tests, JUnit, @Nested, MakeMe builders, @Transactional, parameterized tests. Use when writing or changing backend JUnit tests.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

cli

免费

related to donut CLI. Use when working on donut-cli.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

Set up and run the Donut dev environment on Cursor Cloud VM (no Nix). Use when working on Cloud VM, running tests on Cloud VM, or encountering command errors related to missing nix. Triggers on: cloud VM, cloud agent, no nix available, /workspace path.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

Regenerate docs/database-erd.md from MySQL information_schema (Mermaid ERD: foreign keys and key columns). Use after Flyway schema changes, new migrations, or when the diagram is stale. Triggers on: database ERD, database-erd.md, export_database_erd, schema diagram, ER diagram after migration.

中文简介正在准备中。当前显示原文说明。

nerds-odd-e/doughnut502026年10月26日 更新

查看 nerds-odd-e 的全部技能

报告此技能的问题