shipshape

agent-executed · OSPS Baseline inside · policy as YAML

Point a coding agent at AGENTS.md and it audits an existing repository against the OpenSSF OSPS Baseline plus your house policy, or scaffolds a new one to the same standard. The prompt carries no policy: every opinion lives in YAML you can read, diff, and argue with.

Three files

AGENTS.md

The agent prompt: modes, hard rules, phases. Contains no policy.

standard.yml

Pinned Baseline version, profiles, tiers, house merge and review rules, required files.

stacks/*.yml

Per-stack rules: what to check in an existing repo, what to generate in a new one.

Install

# once, as a Claude Code plugin
/plugin marketplace add dcotelo/shipshape
/plugin install shipshape@shipshape

# then, in any repository
/shipshape audit
/shipshape scaffold

What a run does

  1. Detects the mode. A repository with commits is audited; an empty one is scaffolded. Announced, never assumed.
  2. Resolves profile, tier, and stacks. The profile is asked every run and never inferred from current visibility.
  3. Fetches the pinned Baseline. Control IDs come from the versioned checklist, verbatim, never from model memory.
  4. Evaluates with evidence. Every control records pass, fail, n/a, or not_run alongside the file, setting, or command output that decided it.
  5. Reports, then stops. In audit mode nothing changes until you pick the fixes, and the report never lands in the repository tree.

Ask for a dry run and the guarantee is stronger: it reads the tree, the API, and the checklist, then writes nothing to the repository or to GitHub. No file in the tree, no git command that writes, no API call that is not a read. The ruleset it would apply is printed rather than sent, and the report goes to stdout unless you ask for a file, which lands outside the repository. That is how you point it at a mature repository without risking one.

What it will not do