Plan Optimizer

An agent skill that turns a vague idea, rough plan, strategy, or prompt into a clear, stress-tested, actionable version. It scores the input, closes intent gaps, offers real alternatives, runs an adversarial audit, and returns a ready-to-use artifact.

Works on product features, research directions, business plans, agent instructions, roadmaps, and one-off prompts.

What it does

1. Classify and score — idea / plan / prompt / strategy, rated 1–10 on clarity, specificity, completeness, actionability, and risk awareness.
2. Extract intent — asks only the questions that change the outcome: success, audience, constraints, non-goals, failure modes.
3. Generate approaches — two or three genuine alternatives with trade-offs, plus a recommendation; waits for a pick.
4. Adversarial audit — hunts contradictions, untestable acceptance criteria, missing edge cases, oversized scope, unsafe assumptions, and unguarded irreversible actions. Each finding gets severity, evidence, and a patch.
5. Deliver — analysis, optimized artifact, open questions, next step.

It also works in quick mode: stop after the audit and skip the rewrite.

Install

The skill is a folder with a SKILL.md. It ships as two archives in this repo; extract one, then copy the folder into your agent's skills directory:

tar -xf plan-optimizer-skill-aio.tar -C /tmp

# OpenCode
mkdir -p ~/.config/opencode/skills && cp -r /tmp/plan-optimizer ~/.config/opencode/skills/

# Claude Code
mkdir -p ~/.claude/skills && cp -r /tmp/plan-optimizer ~/.claude/skills/

Or install the single-file package directly:

unzip plan-optimizer.skill -d ~/.config/opencode/skills/

Restart the agent. The skill triggers on phrases like "optimize this plan", "critique this approach", "improve this idea", or whenever a plan or prompt needs structure.

Files

Path	Contents
SKILL.md	Trigger description, five-step workflow, output template, heuristics
references/scoring-and-frameworks.md	Scoring rubric and framework selection
references/audit-checklist.md	Adversarial audit checklist
references/examples.md	Worked examples
evals/evals.json, evals/triggers.json	Evaluation cases and trigger tests (in the .tar)
plan-optimizer.skill	Zip-packaged skill
plan-optimizer-skill-aio.tar	Full source bundle

Example

You:  Optimize this plan: build a Chrome extension that summarizes tabs.
Agent: Scores (clarity 4, specificity 3, ...) → asks what "summarize" means and
       who it's for → offers 3 approaches (local model, API, hybrid) →
       finds 6 audit issues (permissions, cost, privacy, ...) →
       returns a rewritten plan with success criteria and verification steps.

Design notes

• Never invents domain facts about the user; uses [you fill this in: ...] placeholders instead.
• Prefers the existing structure when the input is already decent.
• Optimized output is shorter and denser than the input unless the input was incomplete.
• For agent-facing plans it adds explicit verification steps and stop conditions.

License

MIT © 2026 TonyD2F — see LICENSE.