Askit build agents md

Skill product-on-purpose/agent-skills-toolkit/skills/askit-build-agents-md

Toolkit and Standard for building, grading, and scaling cross-agent skill libraries (Claude Code + Codex) to a tiered Bronze/Silver/Gold quality bar.

Install

npx -y skills add product-on-purpose/agent-skills-toolkit --skill askit-build-agents-md

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Creates and improves a plugin's AGENTS.md (the agent navigation and instructions entrypoint) to the Advanced Skill Library Standard. Use when you need to author or sync AGENTS.md, align it with the component index, or trim an overgrown one to essential, mostly-positive guidance.

SKILL.md

2.5 KB, as published. Nobody here has run it

askit-build-agents-md

Purpose

Author and keep current a plugin's AGENTS.md: the project-level agent instructions and navigation entrypoint, the first thing an agent reads (Standard sec 3.10). It is the single most portable cross-tool surface (Claude, Codex, Gemini, Copilot, and Cursor all read AGENTS.md). Two modes: create authors a tight AGENTS.md; improve syncs it to the component index and trims it. Authoring depth is in references/authoring-agents-md.md. Follows the shared builder contract (../../docs/reference/builder-pattern.md).

When to use

When the user asks to author, write, sync, or trim AGENTS.md (or CLAUDE.md), or when components were added or removed and the navigation is stale.

create mode

Brief interview: what the project is, the conventions (runtime, build/test/lint), and the boundaries. Skip if in context.
Author a TIGHT AGENTS.md: what-this-is, current-state (read before assuming capabilities), conventions, build/test/lint, where-to-look. Keep it minimal and essential-only, and prefer positive instructions over long "do not" lists. Verbose, generated-looking context files measurably reduce agent task success and raise cost, so brevity is a feature, not a shortcut.
Keep the component list aligned with library.json and the index. node scripts/generators/sync-agents-md.mjs renders the component section from the manifest so it does not drift.
Evaluate (node scripts/evaluate.mjs . --json); AGENTS.md presence is checked by U2 (anatomy), and the description and instruction-budget rules apply.

improve mode

Run evaluate and read findings; sync the component list to disk.
Trim: remove redundancy and non-essential prohibitions; judge each change by whether it helps an agent act correctly, not by length.
Re-run to confirm.

Scope

AGENTS.md is Universal-tier and the cheapest credible cross-tool surface. Keep it brief and current; the generated component section (sync-agents-md) prevents drift from the manifest. A CLAUDE.md is a Claude-specific sibling for the same purpose when a project needs Claude-only guidance.