dev-do

Executes implementation plans phase by phase with verified local commits.

Updated Mar 9, 2026
One-click install
npx skills add https://github.com/GinoCanessa/dotnet-fhir-packages --skill dev-do-ginocanessa
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dev-do
Source: https://github.com/GinoCanessa/dotnet-fhir-packages/tree/main/.github/skills/dev-do
Command: npx skills add https://github.com/GinoCanessa/dotnet-fhir-packages --skill dev-do-ginocanessa

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a written implementation plan into actual code changes is slow and error-prone when done manually: phases get skipped, commits mix unrelated files, and verification steps are forgotten. This Skill drives a plan.md file to completion autonomously, enforcing clean working-tree gates, per-phase verification, and one scoped commit per phase. ## Core Features & Use Cases - Plan-driven execution: Reads a plan.md produced by dev-plan (by full path or slot number), then runs every Pending phase back-to-back, updating statuses and a Progress Log as it goes. - Verified commits: Stages only phase-owned paths, commits with git commit --only, and validates parent, tree, and changed-path identity after every commit. Never pushes and never opens a PR. - Sub-agent orchestration: Delegates implementation to dev-implementer agents and builds/tests to task agents, capped by a configurable max_subagents concurrency limit. - Recovery mode: Re-invocation after a Blocked phase or interruption reconciles durable Progress Log evidence instead of blindly rerunning work. - Use Case: After planning a multi-phase refactor with dev-plan, invoke this Skill with the slot number and let it implement, test, and commit each phase locally while you review checkpoints. ## Quick Start Run the dev-do skill with slot 2 to execute today's plan file, implementing each phase, running its verification commands, and committing locally as phases complete.

Frequently Asked Questions about dev-do

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

FAQPage Schema
How do I execute an implementation plan automatically with an AI agent?

Provide the path to a plan.md file (or a slot number that expands to scratch/<MMDD>-<##>/plan.md) and the Skill runs every Pending phase: implementing code, running verification commands, and committing locally. It stops only on blocked phases, scope-exceeding decisions, or configured checkpoints.

How do I make an AI coding agent commit changes safely per phase?

The Skill stages only the phase's declared owned paths, commits with git commit --only, then verifies the new commit's parent, tree, and changed-path set match the recorded pre-commit values. Any mismatch marks the phase Blocked instead of rewriting history.

Does this Skill push commits or open pull requests?

No. It creates local commits only and explicitly never runs git push or gh pr create. Pushing and opening the PR are delegated to a separate dev-pr-open skill, keeping the execution stage isolated from publishing.

What happens when a phase's tests or build fail during execution?

A failed verification is a stop condition: the Skill debugs within reasonable scope, and if the command cannot be made green it marks the phase and plan Blocked with an actionable reason and yields to the user rather than continuing.

Can I resume an interrupted or blocked plan run?

Yes. Re-invocation enters recovery mode: it requires a clean index, inspects the Progress Log's PENDING entries with recorded HEAD, tree, and changed paths, reconciles any matching commit, and resumes from the last safe gate instead of rerunning completed work.

How many sub-agents can run in parallel during implementation?

The max_subagents parameter caps concurrent sub-agents, defaulting to 3 with a hard upper bound of 8. It is a concurrency cap, not a total cap, so more agents may run sequentially over the life of the task.