home-modules/shared/skills/go-do/SKILL.md
Goal-driven execution. Breaks a task into phases with verification, parallelizes via subagents, and iterates until the outcome is achieved. Triggers: 'go-do', 'goal-driven', 'achieve this', 'make it so'.
npx skillsauth add mccurdyc/nixos-config go-doInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Efficiently and effectively achieve a specified goal through structured decomposition, parallel execution, verification, and iteration.
When activated, gather the following from the user. If the user provides a
brief instruction (e.g., "go-do: make the tests pass"), ask clarifying
questions for any missing pieces. Use the interactive-questionnaire skill
if there are many options to resolve.
| # | Question | Purpose | |---|----------|---------| | 1 | Motivation — Why are you doing this? | PR description, commit messages | | 2 | Outcome — What is the desired final state? | Success criteria | | 3 | Phases — Are there logical stages? | Decomposition | | 4 | Verification — How do we confirm each phase and the final outcome? | Concrete commands/checks | | 5 | Iteration — What's the rebuild/deploy/verify loop? | Worker instructions | | 6 | Guardrails — What MUST or MUST NOT happen? | Hard constraints | | 7 | Cross-references — Docs, files, or examples to consult? | Context for workers |
If the user's intent is clear enough to infer answers, propose them and confirm rather than asking redundant questions.
After intake, write a .go-do.md file in the project root with the
structured answers. This file is the single source of truth for all
subagents. Pass its path to every worker so they have full context.
# Goal Contract
## Motivation
<why>
## Outcome
<desired final state — specific, verifiable>
## Phases
1. <phase> — done when: <verification command/criteria>
2. <phase> — done when: <verification command/criteria>
## Iteration Loop
- Rebuild: <command>
- Deploy: <command>
- Verify: <command>
## Guardrails
- MUST: <requirement>
- MUST NOT: <prohibition>
## Cross-References
- <path or URL>
The orchestrator breaks the goal into phases. For each phase:
tmux-delegate..go-do.mdAfter each worker reports done, the orchestrator runs the phase verification independently. Trust but verify — don't take the worker's word for it.
For the final outcome, run the outcome verification (which may be a superset of phase checks — e.g., full test suite, end-to-end test, build from clean).
When verification fails:
Attempt 1-3: Provide the failure output as context to the next worker attempt. Each attempt must show measurable progress (fewer errors, different error, partial fix). If an attempt produces the exact same failure with no progress, count it as wasted and escalate sooner.
After 3 attempts without clear progress: STOP. Do not continue iterating. Iteration without progress is a signal that context is missing. Ask the user:
git-worktree skill for isolation.The orchestrator decides during planning and tells the user the strategy.
Before starting phase 1, verify:
If any pre-flight check fails, resolve it or ask the user before proceeding.
.go-do.md before starting work..go-do.md after the goal is achieved and committed.When the final outcome verification passes:
.go-do.md.tools
Do work in an isolated git worktree instead of switching branches. Use when creating a branch and opening a PR so the user's working directory is never disturbed. Triggers: 'create a branch', 'open a PR', 'make a change on a new branch'.
development
Search the web, fetch web page content, or search GitHub issues/PRs/repos. Use when you need current information not available locally.
tools
Start an interactive questionnaire when there are more than 5 options or bullet points that need to be addressed. Uses the ask_user tool to walk through selections interactively instead of dumping a wall of text.
tools
Browse the web headlessly using Chrome DevTools Protocol. Use when you need to interact with web pages - click buttons, fill forms, navigate, take screenshots, or extract dynamic content that requires JavaScript.