skill-reverse-engineer

Treat any visible skill artifact as a designed system — read out its trigger logic, workflow, output contract, risks, and reusable creation formula, keeping evidence and inference cleanly separated.

When to use it

Diagnose a skill quickly:

You're staring at a SKILL.md and want a fast read on what it's really doing, why it triggers where it triggers, and whether it's any good — no oversized report.

Analyze a skill in depth:

You want to understand end to end how a skill was built — trigger mechanism, workflow steps, reused design patterns, and the creation formula behind it.

Find improvement areas:

The skill is weak — vague trigger, missing output contract, overlong body — and you want a targeted patch list or a fully rewritten SKILL.md.

Test when it triggers:

You want to know whether the skill actually fires on the requests you care about — should-trigger / should-not / ambiguous / edge queries with reasons.

Compare two skills:

You hand over multiple skills and want the shared formula, the differences, and a reusable template extracted.

Prepare for publishing:

Turn a raw prompt workflow into a marketplace-ready skill — positioning, folder structure, SKILL.md, tests, release checklist.

Won't take:

Bootstrapping a brand-new skill from a blank page → skill-create-workflow; naming a skill's slug / display name → skill-domain-framing / skill-name-generation; deciding whether raw material is worth crystallizing into a skill → skill-content-fit; installing / listing / running an existing skill → runtime CLI, not this skill.

What it produces

Evidence-labeled analysis, mode-sized to the request. It rewrites a full SKILL.md only when you ask.

• Mode first: Picks Quick / Full / Audit / Trigger-test / Package / Compare before writing anything, so a light question doesn't come back as a full report.
• Evidence discipline: Every material claim is tagged — "from the visible content" / "reasonable to infer" / "cannot be determined because" / "to verify this, test". Never claims access to hidden system prompts, private routing, or marketplace ranking logic.
• Structural diagnosis: Trigger logic, workflow steps, output contract, resource strategy, risks and failure modes.
• Trigger benchmark: When trigger reliability matters, produces 10 should-trigger / 10 should-not / 5 ambiguous / 5 edge queries, each with expected activation and reason.
• Quality scoring: 15 dimensions (intent clarity, trigger precision / recall, workflow completeness, output consistency, safety, portability, context efficiency, marketplace readiness, ...) plus an overall rating.
• Creation formula: 16-point reusable formula summarizing how the skill was likely built — target user, repeated task, workflow pattern, trigger strategy, resource strategy, output contract.

Boundaries

Only reads what the user pastes or points to; never claims platform-internal knowledge (hidden system prompts, marketplace ranking, private routing); missing evidence is called out, not fabricated.

Adjacent skills:

Task	Route to
Bootstrap a brand-new skill from a blank page	skill-create-workflow
Name a skill's slug / display name	skill-domain-framing / skill-name-generation
Decide whether raw material is worth crystallizing into a skill	skill-content-fit

Subtle edges:

• Partial artifacts are analyzed with missing sections explicitly labeled — never fabricates closure
• A fully rewritten SKILL.md is emitted only when the user asks for it, or when patch-style advice is too weak to help