implement-task

Implements planned tasks from task packets with strict plan adherence and reviewable increments.

2|Updated Apr 29, 2026
One-click install
npx skills add https://github.com/siarhei-belavus/agent-public --skill implement-task-siarhei-belavus
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: implement-task
Source: https://github.com/siarhei-belavus/agent-public/tree/main/skills/atelier-workflow/implement-task
Command: npx skills add https://github.com/siarhei-belavus/agent-public --skill implement-task-siarhei-belavus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It prevents silent plan drift, duplicated logic, and unmaintainable code by enforcing a disciplined implementation workflow driven by a written plan in .plans/<task-slug>/. ## Core Features & Use Cases - Plan-as-contract execution: Refuses to code without PLAN.md, stops on blockers, and records human-approved changes in AMENDMENTS.md before continuing. - Ownership and reuse enforcement: Prefers extending existing owner modules over creating parallel helpers, and prohibits backward-compatibility shims unless explicitly required. - Review-fix loop and handoff: Treats routed review findings as immediate work, validates the risky path, commits each pass with a Conventional Commit, and reports back to the requester. - Use Case: A lead agent assigns a planned refactoring task; the skill repairs packet drift, implements the change in reviewable increments, runs targeted validation, commits, and hands off with updated packet artifacts. ## Quick Start Implement the task defined in the current task packet under .plans, following PLAN.md strictly and committing the result before reporting back.

Frequently Asked Questions about implement-task

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I implement a planned task without deviating from the plan?▼

Treat PLAN.md as a contract and follow it strictly. If a blocker or contradiction makes the plan unsafe, stop, report options with pros and cons, let the human decide, and record the decision in AMENDMENTS.md before continuing.

How should AI agents handle review findings during implementation?▼

Treat routed P#, PS#, or M# findings as executable work and start the fix pass immediately in the same turn. After fixing, rerun the smallest validation proving the fix, update packet artifacts, commit the fix pass, and report results.

When is backward compatibility allowed in a code change?▼

Backward compatibility is prohibited by default. Aliases, shims, and fallback behavior are only allowed when the human explicitly requires them, and unresolved external-boundary compatibility questions must be escalated before proceeding.

What happens if the plan seems suboptimal during implementation?▼

Stay on plan and do not silently improve it. Leave a REVIEW_NOTE at the relevant point describing the better option and why it may be preferable; the note is advisory and does not authorize scope or design changes.

Why avoid creating a new helper instead of reusing existing code?▼

Creating a parallel same-purpose helper, service, or module counts as plan drift and raises maintenance cost. The policy requires identifying the existing owner of each behavior and extending or tidying it so one authoritative implementation path remains.