implement-task
Use when a task — a slice from a slices file — needs building. The user asks to "implement task 3", "implement slice 3", "build the next task", or "do the next slice". Also when implementation keeps wandering past its slice's bound, or slices get coded but the slices file never reflects it.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- implement-task
- license
- MIT
- metadata
- { "summary": "Implements one task from a slices file to its demo, test-first for behavior changes, and records completion before reporting.", "category": "Building" }
- description
- Use when a task — a slice from a slices file — needs building. The user asks to "implement task 3", "implement slice 3", "build the next task", or "do the next slice". Also when implementation keeps wandering past its slice's bound, or slices get coded but the slices file never reflects it.
02 资源文件 12
- 其他evals/cases.jsonl7.4 KB
- 参考文档evals/fixtures/notifications-slice/docs/specs/2026-07-28-notifications-slices.md1.4 KB
- 参考文档evals/fixtures/notifications-slice/docs/specs/2026-07-28-notifications.md486 B
- 参考文档evals/fixtures/notifications-slice/package.json138 B
- 参考文档evals/fixtures/notifications-slice/README.md1018 B
- 脚本evals/fixtures/notifications-slice/src/notifications.js276 B
- 脚本evals/fixtures/notifications-slice/test/notifications.test.js269 B
- 参考文档evals/results/README.md3.4 KB
- 参考文档evals/results/tier2-2026-08-01-sessions.md38.1 KB
- 参考文档evals/results/tier2-2026-08-01.md4.9 KB
- 其他evals/retired-cases.jsonl466 B
- 参考文档evals/skill-brief.md4.8 KB
03 SKILL.md 正文
Implement Task
Build one bounded task to its written acceptance criterion. For changed production behavior, observe a valid failing check before implementation. Record completion before reporting it.
Scope
Use for one named task, the next unblocked task, or a task whose completion record has fallen behind its code.
Read the slices file and linked spec. Use the named task, otherwise the first unblocked unfinished task. A delivery coordinator may invoke this workflow again for the next task covered by the user's request.
For several tasks, use deliver-feature when installed. Otherwise repeat this workflow in dependency order within the authorized scope. If no plan exists, create one only when the work needs decomposition; build a small, clear task directly.
Workflow
-
Read the contract. Check the spec's confirmation and the slices file's
status. When either is open, ask for that confirmation before building. Then check the task's bound, demo, blockers, and the current worktree. Honor repository path overrides.When the plan exists only in chat, ask to save it beside the spec as
<spec name>-slices.mdbefore building.Stop dependent work when a blocker is incomplete. State the target task and bound. Preserve user-authored and pre-existing work.
Done when the task is confirmed, unblocked, and bounded.
-
Encode the acceptance check. For changed production behavior, write the test first. Run it and observe the intended behavioral assertion fail. Use
tddwhen installed; otherwise retain that red-first order.For documentation, formatting, or mechanical changes without a behavior effect, use proportionate existing checks. Record why a failing behavior test is inapplicable. Repair broken test setup instead of counting it as red.
Done when valid red is observed or a non-behavioral verification choice is recorded.
-
Implement inside the bound. Make the smallest change that satisfies the unchanged check. Treat
Boundas permission;Layersdescribes coverage only. Record adjacent work for a later task.When evidence contradicts the plan, record a correction before code relies on it. Clarify or shrink the active bound. Put expanded scope in a new task; ask before changing confirmed behavior or scope. Leave incomplete work unticked.
Done when the check passes and every changed surface belongs to the active task.
-
Run the acceptance criterion. Execute the written demo literally. Run focused tests and the affected package or subsystem suite. Run the full suite when proportionate and available; name unrun checks.
For scripts, configurations, parsers, or rule documents, use
verify-workwhen installed. Pass along only existing in-scope fix authority. Otherwise execute applicable adversarial cases against the changed promise, including empty, malformed, and boundary inputs when meaningful.If the exact demo cannot run, use the closest executable observation and record the weaker evidence. Leave the task incomplete when its acceptance promise is unproven.
Done when the acceptance criterion and applicable surrounding checks pass.
-
Record before reporting. Tick the task only after verification. Add
Done: <date>: <one-line outcome>. Clear or supersede whole-feature verification invalidated by these edits. With no slices file, put the dated outcome in the report instead.Report the task, observed demo, commands, changed files, and remaining work. Commit only when authorized by the request or binding repository workflow.
Done when the completion record matches the demonstrated result.
Example
Task 2 adds mark-read behavior for an existing notification list. Its bound names two endpoints, the list interaction, and focused tests. Observe a failing unread-count assertion first. Implement the two endpoints and interaction. Run the written “badge 1 → mark-read → badge 0” demo. Tick task 2 after checks pass; leave an unrequested event source for its own task.
Pressure checks
| Shortcut | Required action |
|---|---|
| “The neighboring handler is only two lines.” | Keep it outside this task unless the bound includes it. |
| “Tests are green, so the demo is obvious.” | Execute the written acceptance criterion. |
| “Write the test after this small change.” | Observe valid red first for production behavior. |
| “I'll tick the plan after the report.” | Update the completion record first. |
| “Urgency lets the bound expand.” | Correct the plan and obtain any missing scope decision. |
Follow the workflow silently. When a step is genuinely blocked, name the evidence, use the closest valid alternative, and state its limitation.
Verification
- Changed production behavior has valid red-to-green evidence.
- The patch stays inside the confirmed task bound.
- The demo and applicable checks actually ran.
- Unproven acceptance leaves the task unticked.
- The plan records completion before the final report.
Build one bounded task, demonstrate its result, and record the evidence before claiming completion.
04 同仓库更多 Skills
write-agent-instructions
Make the smallest evidence-backed change to persistent project instructions and test downstream behavior.
verify-work
Use when a finished, runnable artifact (a script, config, parser, prompt, or CLI, not a diff under review) must survive hostile input before shipping: the user asks to "verify this script", "find the loopholes", "test this properly against bad input", or asks "is this ready to ship" about one. Also when a shipped artifact failed on an input nobody tried, or when verification so far is a happy-path smoke test.
understand-codebase
Use when the user asks to "understand this codebase", "how does this feature work in this repo?", "trace this feature through the code", or needs onboarding, an architecture explanation, or answers about unfamiliar code. Also use for read-only code-path questions before a change.
tdd
Use when writing or changing production behavior: implementing a feature, fixing a bug whose cause is already known, or altering stored, transformed, migrated, serialized, or externally transmitted data. Also use when a user says "use TDD", "test first", or "red-green-refactor", and when code is being written before a failing test. For a failure whose cause is unknown, diagnose first.
simplify-code
Use when the user asks to "simplify this code", "remove the fluff", or "clean up this over-engineered module", or wants working code that is bloated, duplicated, over-abstracted, or dependency-heavy reduced without changing behavior.
review-session
Review an agent session for evidence-backed friction and the smallest measurable remedies.