AGENT SKILL

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 "work through the slices". 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
所属仓库
adriancodes/skills
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 slice to its demo criterion, starting with a failing test and ticking the slices file before reporting completion.", "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 "work through the slices". 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 2

03 SKILL.md 正文

Implement Task

Overview

Implement one recorded slice per session. Stay inside its bound. Completion requires a successful demo and a ticked slices file. Green tests alone are insufficient.

When to Use

  • "implement slice N", "build the next slice", "keep working through the slices"
  • A slices file exists (from create-tasks or hand-written) with an unblocked, unticked slice
  • Slices keep getting coded while the slices file silently rots

Do Not Use When

  • No slices file exists: use create-tasks when installed. Otherwise write demoable, session-sized pieces. Build directly only when the work truly fits one session.
  • More than one slice is wanted: one invocation, one slice; the next gets a fresh session
  • The slice needs replanning: change the slices file first. Reopen spec decisions through a new superseding entry with the user's approval.

Required Context

  • The slices file and linked spec, read in full before code.
  • Each slice needs Layers, Bound, Demo, and Blocked-by fields.
  • The file needs status:, ## Confirmation, and ## Verification.
  • The target slice: the one named, else the first unblocked unticked slice
  • The repo's test and build commands

Workflow

  1. Anchor. Read the slices file and linked spec. Check every blocker. If one is open, stop and name it. Restate the slice's layers, demo, and bound.

    Done when the restatement is in the conversation.

  2. Encode the demo as a failing check. Write a test or executable check before implementation. Make it express the demo criterion. Run it and observe the intended failure. Follow tdd when installed. Without it, preserve the same red-first order.

    Done when the check fails for the right reason.

  3. Build inside the bound. Treat Bound as authoritative. Layers describes coverage, not permission to touch every file. Record adjacent work for a later slice.

    When reality contradicts the bound, stop. Correct the slices file before code. The correction may shrink or clarify this slice. Put new scope in a fresh slice. If the demo needs that new scope, narrow the demo or stop at a clean seam. Urgency never expands the active slice.

    Done when the check passes and every changed surface is inside Bound.

  4. Run the demo literally. Execute the written criterion. Then run focused tests and the full suite once.

    Done when the demo is observed and both test levels pass.

  5. Verify hostile surfaces. For scripts, configs, parsers, prompts, or similar artifacts, run verify-work when installed. Pass explicit authority for fixes inside the bound. Stop before any fix that expands it. Without verify-work, execute at least 3 hostile cases: empty, malformed, and boundary.

    Done when scope-bound fixes pass re-attack. Otherwise leave the slice unticked and record residual findings.

  6. Tick before telling. After steps 4 and 5 pass, tick the slice. Add a dated one-line outcome note. Report the slice, demo, suites, hostile checks, changed files, and later notes. Commit when the repository commits per work unit.

    Done when the file shows the tick and the report links it. Failed verification leaves the slice unticked.

Example: the two orderings

Baseline order (forbidden): endpoints → UI wiring → test written against finished code → passes first run → "done, tests green." This skill's order: failing badge test → implementation → observed demo → hostile checks → file tick → report.

Common Rationalizations

The usual ways implementation escapes its slice's bound.

Excuse Reality
"While I'm in this file anyway, I'll wire the other event source: two lines" Those are the next slice's two lines. Note them there; the bound holds.
"I'll add mark-all-read too, it's basically the same handler" "Basically the same" turns a two-endpoint slice into four. The bound names two.
"The demo is obvious from the code; no need to click through" Obvious demos fail on the seam nobody typed: run it, watch it.
"Tests are green, so the demo criterion is effectively satisfied" The test is the criterion encoded; the demo is the criterion executed. Both, in that order.
"I'll tick the slices file at the end" "The end" is after the report, which is never. Tick before telling.

Success Criteria

  • The demo-criterion check existed and failed before implementation code
  • Zero changes outside the slice's bound.
  • Record any bound correction before code relies on it. Never build an expansion in the same session.
  • The demo criterion was literally executed and observed
  • Relevant tests and the full suite passed
  • Hostile-input verification passed or was explicitly inapplicable
  • The slices file is ticked with an outcome note before the final message

Common Mistakes

Mistake Fix
Starting from the slice title alone Read the slices file and the spec it links, in full
Test written after the code Delete it as evidence; red first (step 2)
Bound violation reported as a bonus It's a plan defect: slices file first, then code
"Done" in chat, [ ] in the file The file is the pipeline's state; chat is not

Failure Modes

  • A blocker is unticked: stop and name it; the blocker's slice comes first.
  • The demo cannot run: use the closest executable observation. Label the substitution in the outcome note. Never silently downgrade to "tests pass."
  • The slice exceeds one session: stop at a clean seam. Add the remainder as a new blocked slice. Tick nothing.

Summary

Implement one bounded slice at a time. Observe the failing test first, run the demo, and tick the slices file before reporting completion.

04 同仓库更多 Skills

adriancodes
adriancodes/skills

write-agent-instructions

Make the smallest evidence-backed change to persistent project instructions and test downstream behavior.

★ 1
adriancodes
adriancodes/skills

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.

★ 1
adriancodes
adriancodes/skills

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.

★ 1
adriancodes
adriancodes/skills

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.

★ 1
adriancodes
adriancodes/skills

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.

★ 1
adriancodes
adriancodes/skills

review-session

Review an agent session for evidence-backed friction and the smallest measurable remedies.

★ 1