peakspec/marshal/optimize-goal
Claude Code and agent skill by peakspec/marshal
- community
Checking that the SKILL.md file is still in the repository.
Use when the user wants to review, critique, tighten, or enhance an existing goal.md file (a written spec/prompt authored for a /goal command instead of typed into chat) before it gets executed — including requests like "review this goal.md", "enhance my goal.md", "optimize this goal", or sharing a goal.md file and asking for feedback before running it.
Install
npx vskill@latest install peakspec/marshal/optimize-goalSKILL.md
Optimize Goal
Overview
A goal.md is a long-form instruction the user writes to a file instead of typing into chat, meant to be fed to Claude Code's `/goal` command. /goal uses that same text twice: once as the directive for the first turn, and then every turn after, a separate transcript-only judge model re-checks it against the conversation to decide whether to stop. Whatever is ambiguous, unmeasurable, or unverifiable in the file passes straight through into both of those — there's no chat back-and-forth to catch it later. This skill reviews an existing goal.md against a checklist tuned to both general prompt clarity and /goal's actual mechanics, clarifies the gaps with the user, and only then rewrites the file.
Not for: helping someone compose a goal.md from scratch when no draft exists yet — that's superpowers:brainstorming territory. This skill starts from a file that already exists.
Workflow
- Read the goal.md the user points to.
- Silently audit it against the checklist in
references/best-practices.md. Judge
relevance to this goal.md — don't force a fixed template onto it. A one-paragraph goal doesn't need a Prerequisites section; a multi-repo goal probably does.
- Ask clarifying questions one at a time, ordered by impact (biggest ambiguity or
highest blast-radius gap first — e.g. missing success criteria before missing formatting preferences). Prefer multiple-choice phrasing when the gap has a natural small set of answers.
- Draft the enhanced version once questions are answered. Preserve the user's own
structure, section names, and voice — tighten and fill gaps, don't impose a rewrite they didn't ask for.
- Show the draft (or a diff against the original) and get explicit approval.
- Only after approval, overwrite goal.md in place. Never write the file before the
user has signed off on the content.
Checklist quick reference
Full detail with examples in references/best-practices.md. Categories to scan for on every goal.md, applied contextually:
| Category | What's being checked | |
|---|---|---|
| Task clarity | Would a colleague with no context execute this correctly on the first read? | |
| Context / why | Is the motivation behind non-obvious constraints stated? | |
| Success criteria | Is "done" a measurable end state with a stated check the transcript-only evaluator can actually see Claude demonstrate — not just described? | |
| Bound the loop | Is there a turn/time stop clause, so an unverifiable or open-ended condition can't run forever? | |
| Non-goals / scope | Is what's explicitly out of scope stated, to prevent scope creep? | |
| Resources | Are the inputs, docs, or repos the task depends on named and locatable? | |
| Guardrails | Are irreversible/high-blast-radius actions flagged for confirmation? | |
| Execution order | Are steps that must happen in sequence actually ordered? | |
| Prerequisites | Is setup/access needed before starting called out? | |
| Output format | Is the shape of the deliverable (file, PR, report, message) explicit? | |
Fits /goal | If this file feeds /goal directly, does the condition portion fit under the 4,000-char cap? | |
| Ambiguity scan | Any line readable two different ways? |
Common mistakes
- Asking everything at once. Dumping five clarifying questions in one message gets
shallow answers. One at a time, highest-impact gap first.
- Rewriting before approval. Draft, show, wait for sign-off, then write the file.
- Enforcing a rigid template. The checklist is a lens, not a form to fill in — a
short goal.md that's already clear doesn't need every category addressed explicitly.
- Treating "no file yet" as in scope. If the user has only an idea and no draft,
point them to superpowers:brainstorming instead of improvising a first draft here.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/optimize-goal/SKILL.md
Popularity
- Repository stars
- 50
- Repository forks
- 13
- 7-day trend score
- 20
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/peakspec/marshal/optimize-goal)