Premium skill

Write agent prompts you can defend

Rewrites the prompts your agent actually runs on: a repo-wide audit of contradictions and token waste, prompts as versioned code, and before and after evals proving every change.

Install with Premium View pricing

Plan: Premium
Access: Gated
Install: Tokenized
Authored files: 9

Install with skills CLIPremium skills install through a short-lived private link.

$ npx skills add … --skill write-agent-prompts

Open Premium install

Filesystem

Browse the authored file tree — tools, references, and evals included. Source unlocks with Premium.

SKILL.mdmarkdown

Unlock skill source

Browse every authored SKILL.md with premium.

Install with Premium

File tree for write-agent-prompts stays visible so you can see the pack before unlocking.

File

What you get

Effective prompt inventory

Treat system text, tool descriptions, schemas, and examples as one document. Audit contradictions and dead instructions before rewriting.

Prompts as versioned code

Extract scattered strings into owned modules with typed parameters, then prove equivalence before craft rewrites.

Measured before and after

Establish a golden baseline first. Keep only prompt slices that improve the gate; revert the rest.

Untrusted content framed

Mark retrieval, tool results, and uploads as data, never instructions, with a trust boundary the model can follow.

Where this skill lands

Measured prompts, not vibed prompts: inventory, audit, rewrite, and prove every change.

An agent that ignored half its instructions after years of prompt accretion.
A 4,000-token monolith pulled into reviewable prompt modules with no behavior change.
Wrong-tool calls fixed by clearer descriptions, not a silent API redesign.
Few-shot examples chosen from real traces and cut when they did not earn their tokens.

Ship prompt changes with evidence

Available with Premium. The private install link expires; the downloaded skill does not. Generate a fresh link when you want to update it.