sddl-executor

Executes one approved plan stage per invocation with logging and state tracking.

Updated Apr 2, 2026
One-click install
npx skills add https://github.com/nico0695/ai-tools --skill sddl-executor-nico0695
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sddl-executor
Source: https://github.com/nico0695/ai-tools/tree/main/sdd/sdd-lite/skills/sddl-executor
Command: npx skills add https://github.com/nico0695/ai-tools --skill sddl-executor-nico0695

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? It prevents uncontrolled, all-at-once implementation by executing exactly one approved stage from plan.md at a time, preserving scope discipline, explicit user approval, and resumable traceability across a spec-driven development workflow. ## Core Features & Use Cases - Stage-Scoped Execution: Runs a single planned stage from plan.md per invocation and never auto-continues into the next stage. - Stop Rules: Halts on contradiction with approved artifacts, scope drift, or blast-radius expansion instead of silently widening the change. - Resumable Ledger: Maintains execution-log.md and state.yaml so work can resume with full history of approvals, changed files, quick checks, and blockers. - Use Case: After a plan is approved in an sdd-lite change workflow, invoke this stage to implement only stage 2 of the plan, run proportionate quick checks, log the outcome, and recommend QA review before the next stage begins. ## Quick Start Execute the next approved stage from plan.md for my current sdd-lite change and update execution-log.md and state.yaml with the results.

Frequently Asked Questions about sddl-executor

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

FAQPage Schema
How do I execute a plan one stage at a time with approval gates?▼

Use a stage-scoped executor that reads the Stage Plan table in plan.md and runs exactly one stage per invocation. Each stage requires explicit user approval before starting, and the run stops automatically after completion rather than continuing to the next stage.

How to keep implementation work resumable across sessions?▼

Maintain an execution-log.md ledger that appends an entry per attempted or completed stage, recording approvals, changed files, quick checks, and blockers. Pair it with a state.yaml file tracking current stage, lifecycle status, and next action so work can resume safely.

What happens when execution drifts from the approved plan?▼

The stage stops before widening scope when drift is detected, such as work requiring deliverables not in plan.md. The drift is recorded in execution-log.md and routed back to the user for approval, replanning, or clarification.

Can the executor run git commands like commit or stash?▼

No. The executor performs no git side effects: no commits, stashes, rebases, or even git add. Staging files belongs to the orchestrator and only on explicit user request.

When should QA review happen between implementation stages?▼

QA review is recommended when a completed stage touched code with non-trivial blast radius, reached a meaningful checkpoint, or surfaced warnings in quick checks. The executor records the recommendation in execution-log.md and state.yaml but never auto-runs the review.