Skip to content
Agentmart
Submit a tool

implement-task

Implement a planned task from `.plans/<task-slug>/` with strict plan adherence, explicit amendments, and no direct writes to tracked `AGENTS.md`.

Skill by siarhei-belavus

IndexedMIT★ 2
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/siarhei-belavus-implement-task

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Implement a planned task from `.plans/<task-slug>/` with strict plan adherence, explicit amendments, and no direct writes to tracked `AGENTS.md`.

Use cases: productivity

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/implement-task

What's inside

Skill name
implement-task
Description
Implement a planned task from `.plans/<task-slug>/` with strict plan adherence, explicit amendments, and no direct writes to tracked `AGENTS.md`.
License
MIT
Instructions
213 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# implement-task

Implement changes so they are correct, reviewable, and maintainable. Optimize for small safe increments, clear ownership, explicit validation, and low future change cost.

Treat `PLAN.md` as a contract. Do not improvise around it silently.

## Read first

- `../references/task-packet-contract.md`
- `../references/persistent-artifacts-contract.md`
- `../references/review-phases-contract.md`
- `../references/compatibility-policy.md`
- `../references/ownership-and-reuse-policy.md`
- `../references/final-state-authoring-policy.md`
- `../references/initiative-workflow-contract.md` when implementing an initiative child packet
- `../../references/communication-mode.md`
- relevant tracked `AGENTS.md` chain
- applicable `AGENTS.override.md` only if local execution constraints matter
- current task packet: `INDEX.md`, `PLAN.md`, `AMENDMENTS.md`, `ARTIFACT_CANDIDATES.md`

## Implementation workflow

1. Establish scope before coding.
   - Confirm the task, desired behavior, constraints, and nearby boundaries.
   - Refuse to start without `PLAN.md`.
   - Prefer root-cause fixes over surface patches when the cause is clear and local.
2. Repair packet drift before coding.
   - If `AMENDMENTS.md` or `ARTIFACT_CANDIDATES.md` is missing, create the missing placeholder before continuing.
   - Note the packet repair in `INDEX.md`.
3. Identify the true change surface.
   - Find the owner modules, contracts, state transitions, persistence paths, tests, and tracked `AGENTS.md` constraints that define the behavior.
   - Distinguish direct edits from incidental neighbors.
   - Distinguish internal boundaries from real external contracts relative to the service.
   - Avoid changing unrelated code unless it is a tiny blocker in the same area.
4. Apply ownership-and-reuse policy.
   - Identify the existing owner for each important behavior or invariant before adding code.
   - Prefer reusing, extending, or slightly refactoring that owner over creating a parallel helper/service/adapter/store/module/type.
   - If existing code is close but awkward, do a small tidy-first refactor that makes reuse possible.
   - Do not keep old and new same-purpose implementations alive together.
   - If a genuinely new boundary is required, make the non-overlapping responsibility explicit in code and packet notes.
5. Apply the compatibility policy.
   - Backward compatibility is prohibited by default.
   - Do not add aliases, shims, dual support, fallback behavior, or compatibility wrappers unless the human explicitly required compatibility.
   - If the change touches a real external boundary and compatibility requirements are unresolved, stop and ask the human explicitly before proceeding.
   - For internal-only changes, do not ask and do not preserve compatibility.
6. Implement the chosen target state directly.
   - Move canonical code/docs/contracts toward the steady-state design described in the plan.
   - Remove superseded vocabulary and same-purpose residue inside owner artifacts instead of layering new semantics on top of old names.
   - If a real external boundary requires temporary compatibility behavior, keep the steady-state owner and current contract obvious.
7. Implement in reviewable increments.
   - Follow the plan strictly.
   - Keep the system coherent after each step where possible.
   - Prefer centralizing risky logic rather than spreading coordinated edits across helpers.
   - Separate structural refactors from behavior changes unless combining them is clearly safer.
8. Preserve maintainability while coding.
   - Keep ownership obvious.
   - Minimize new sources of truth.
   - Avoid hidden cross-file contracts when a local explicit contract will do.
   - Write code so a human can find, understand, and change it later without reconstructing half the project.
   - Keep canonical owner artifacts in one vocabulary and one state model.
9. Validate deliberately.
   - Start with the smallest test or check that proves the changed behavior.
   - Expand validation as confidence grows.
   - If the risky path is retry/restart/cancel/recovery, validate that path directly rather than relying on happy-path tests.
10. Hand off cleanly.
   - Keep `AMENDMENTS.md`, `ARTIFACT_CANDIDATES.md`, and `INDEX.md` current enough that both review phases can start cold.
   - Record durable residue candidates when they appear.
   - Commit the implementation pass before reporting back to the lead.
   - Never write tracked `AGENTS.md` or `AGENTS.override.md` directly.

## Plan adherence rule

- Never silently deviate from the agreed plan.
- Unauthorized backward-compatibility work counts as plan drift unless the plan or later human amendment required it explicitly.
- A new same-purpose entity that the plan did not explicitly justify also counts as plan drift.
- Leaving canonical code/docs/contracts in a visible old-plus-new hybrid shape also counts as plan drift when the target design is already known.
- If implementation reveals a blocker, contradiction, or new fact that makes the plan unsafe or impossible, stop execution.
- Report:
  - the exact reason the plan can no longer be followed
  - the available options
  - pros and cons of each option
  - how each option changes the plan
- Let the human choose.
- After the human decides, update `AMENDMENTS.md` before continuing implementation.

## Suboptimal plan rule

- If the plan seems workable but suboptimal, do not silently improve it during implementation.
- Stay on plan.
- Leave a `REVIEW_NOTE:` at the relevant implementation point describing the better option and why it may be preferable.
- Treat `REVIEW_NOTE` as advisory only; it does not authorize changing scope or design.

## Implementation principles

- Make the safest next edit obvious.
- Prefer local reasoning over clever indirection.
- Keep risky logic centralized.
- Leave the code easier to review than before.
- Make canonical owner artifacts read like the current system, not like notes about an in-progress migration.
- Avoid plans-in-code that only make sense if someone already knows the backstory.
- Prefer tidy-first reuse over greenfield duplication.
- Keep one authoritative implementation path per important behavior whenever possible.

## When to slow down

Pause and think harder when:

- the task changes persistence or state ownership
- restart, retry, cancellation, or recovery semantics are involved
- the change introduces a second source of truth
- the easy fix requires scattered coordinated edits
- a test passes, but does not actually prove the risky behavior
- the real code shape disagrees with the plan
- a better design appears mid-implementation but was not approved in the plan
- you are about to add a new same-purpose entity because the current owner is inconvenient to reuse

## Initiative child packets
For initiative-created child packets:
- stop immediately if `PLAN.md` still says `Status: not_authored`

From siarhei-belavus/agent-public, MIT.

More from siarhei-belavus/agent-public

Signals

Stars
2
Forks
0
Open issues
0
Last commit (this folder)
Apr 29, 2026 (6 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from siarhei-belavus

Cite this page

implement-task skill. Agent Mart. https://agentmart.dev/skills/siarhei-belavus-implement-task (last updated 2026-10-24)