agent-ready-environments

The audit scores eight areas an agent needs to finish work with evidence: code structure, static checks, tests, documentation, review, a runnable environment, bounded autonomy, and evaluated guidance. You get a fix order based on the gaps it finds.

For the reasoning behind the audit, read the Agent-ready environments lesson.

Install the skill with the command above. Then ask for the audit you need; the agent selects it when your request matches. For example:

Audit this repository for agent readiness. Read-only, please.

Use it when starting a project, auditing an existing repository, or reviewing instruction files such as AGENTS.md and CLAUDE.md, skills, and review gates. Revisit guidance after a model release to check what still helps.

For a read-only audit, the agent uses CI results and source inspection and lists skipped local runs. Required checks, branch rules, and review history need read access to the code host's API, such as gh api. Without access, those checks stay unverified.

Each area, called a pillar, gets the highest score its verified evidence supports:

A fresh checkout must reach a seeded, healthy app with one command for the environment pillar's runnable check. Pillars that do not apply are marked N/A and excluded from totals. Skipped checks cannot raise a score.

The audit also distinguishes advisory guidance, which the model may decide does not apply, from enforced checks such as hooks, deny rules, and required CI. Rules with no exceptions belong in enforcement.

SKILL.md has the pillar criteria, audit modes, and evidence rules.

• An evidence section and a table of scores, gaps, proposed fixes, and whether each fix is advisory or enforced.
• Fixes ordered first by protection of secrets, protected paths, and irreversible actions, then by durability and cost.
• Unverified checks, what running them needs, and gaps in the skill itself.