Agent Tools · Collection 01

Five checks. One clean entry point.

Audit permissions, resolve instruction scope, review configuration changes, lint hooks, and prove packaged plugins without learning five installation flows.

npx --yes --registry=https://codeberg.org/api/packages/automa-tan/npm/ agent-tools@latest list

Install the addon Release 0.1.2

Pick the question

Five specialists, one dispatcher.

Friendly aliases route to exact published releases. Each specialist stays independently versioned and independently useful.

01 · permissions

PermitLint

Are these coding-agent permissions broader, weaker, or less effective than they look?

agent-tools permissions . --json

02 · instructions

RuleRoute

Which AGENTS.md, CLAUDE.md, imports, and scoped rules actually apply here?

agent-tools rules --target . --harness auto --json

03 · review

HarnessDelta

What capability changed in this working tree, staged diff, or Git range?

agent-tools diff . --staged --json

04 · hooks

HookLint

Will this hook configuration be accepted, skipped, duplicated, or misread?

agent-tools hooks --target . --harness auto --json

05 · packaging

PluginProof

Does the package contain what its Codex and Claude Code manifests promise—without loading the plugin?

agent-tools plugins . --harness both --check

Agent Skill bundle

Six focused skills, with deterministic tools underneath.

One router plus a focused workflow for every checker lets Codex and Claude Code select the narrowest useful check without loading unrelated guidance.

Exact versions

The dispatcher pins every specialist. Run agent-tools list --json to see precisely what will execute.

Static boundaries

The checkers parse the configuration in scope. They do not turn static evidence into claims about a running harness.

One escape hatch

Use the bundle for convenience or install any specialist alone. Nothing requires adopting the rest of the collection.