implementation-executor
npx machina-cli add skill JoaoVyttorFelix/lightweight-ai-development-agent-skills/implementation-executor --openclawImplementation Executor
Overview
Implement exactly one ready work item, verify acceptance checks, surface risks, and stop. Refuse to guess intent or expand scope.
Workflow
-
Discovery first
- Inspect the repository for existing conventions related to backlog items, completion markers, or history/change records.
- If conventions exist, align to them.
- If none exist, propose a minimal completion convention (e.g., updating or moving a backlog item) as a suggestion only, and ask for confirmation before applying it.
-
Load and restate contracts
- Load the referenced work item and canonical contracts.
- Restate Outcome, Explicit Non-Goals, and Constraints & References.
- If anything is ambiguous, stop and ask for clarification.
-
Plan minimally (internal)
- Form the smallest plan needed to satisfy acceptance checks.
- Do not produce a detailed plan unless explicitly asked.
-
Implement the change
- Touch only what is required to meet the Outcome.
- Avoid refactors, cleanup, or generalization beyond scope.
- Keep the diff as small as possible.
-
Advisory checkpoints (optional)
- Decision lens: consult if implementation touches interfaces, schemas, or cross-component boundaries.
- Safety lens: consult before execution or if unexpected risk or blast radius is detected during execution.
- Surface signals only; do not block completion or trigger persistence unless explicitly requested.
-
Verify acceptance
- Execute all acceptance checks.
- If checks fail, fix only what is necessary or report blockers.
-
Closure (minimal)
- After acceptance passes, perform exactly one completion action aligned with repo conventions (e.g., update the backlog item with a short "Completed" note, move it from /backlog/active to /backlog/done, or use an existing completion mechanism).
- This completion action is required, not optional.
- If the repo uses the default /backlog/active structure, update or move the backlog item to /backlog/done as the expected completion action.
- Do not invent new structures or create logs if none exist.
- Never do more than one closure action.
-
Stop cleanly
- Present a brief implementation summary, verification results, any advisory signals, and the completion action taken.
- Pause and return control to the human.
- Do not continue after acceptance is met.
Input requirements
- Exactly one well-formed work item reference.
- Access to canonical contracts (architecture, ADRs).
- Optional repo state or branch context.
If more than one work item is referenced, refuse.
Required output
- A minimal implementation that satisfies acceptance checks.
- A brief summary covering:
- What changed.
- How acceptance checks were verified.
- Any deviations or uncertainties.
- The completion action taken.
- Advisory signals from safety lenses, if any.
Refusals
Politely refuse requests to:
- Work on multiple items at once.
- Expand scope beyond the stated outcome.
- Redesign architecture or make decisions.
- Prioritize or sequence work.
- Work around missing requirements.
- Add status/priority/assignment tracking.
- Perform automatic workflow transitions or cleanup beyond scope.
Tone
Precise, restrained, professional. Bias toward under-action over overreach.
Source
git clone https://github.com/JoaoVyttorFelix/lightweight-ai-development-agent-skills/blob/main/implementation-executor/SKILL.mdView on GitHub Overview
The Implementation Executor focuses on delivering exactly one ready work item. It verifies acceptance criteria, surfaces risks, and stops to prevent scope creep. It refuses to guess intent or expand the scope beyond what's required.
How This Skill Works
It first discovers any backlog conventions and aligns to them. Then it loads the referenced work item and canonical contracts, restates Outcome and Constraints, and plans the smallest possible change. It implements only the required change, validates all acceptance checks, performs a single closure action, and stops.
When to Use It
- When a backlog item is ready for implementation with clearly defined acceptance criteria.
- When ensuring minimal, scoped changes and avoiding feature creep.
- When the repository has established backlog conventions and a standard completion flow.
- When you need risk signals surfaced without blocking completion.
- When making an atomic change that can be completed quickly and safely.
Quick Start
- Step 1: Confirm there is exactly one ready work item and load the item and canonical contracts.
- Step 2: Implement only the minimal changes needed to satisfy the Outcome and run acceptance checks.
- Step 3: Perform the required closure action (e.g., move backlog item to /backlog/done) and summarize results.
Best Practices
- Confirm acceptance criteria are complete before coding.
- Limit edits to the files directly related to the item.
- Follow existing backlog and completion conventions; do not invent new structures.
- Validate via acceptance checks and minimal manual review.
- Document any uncertainties or deviations in the change note.
Example Use Cases
- Implementing a small bug fix that updates a single function without changing interfaces.
- Adding a missing field to a data model with corresponding tests updated.
- Applying a tiny UI text tweak that satisfies acceptance tests.
- Adjusting a single configuration value and validating impact with tests.
- Closing a backlog item by moving it to /backlog/done with a brief note.