Security Slam Agent Skills

Agent Skills for earning Security Slam project badges.

The skills follow the open Agent Skills standard, so they work with any compatible agent, including Claude Code, Codex, GitHub Copilot, Cursor, and Gemini CLI. For Claude Code, this repository is also a plugin marketplace named security-slam that ships them as one plugin, security-slam-skills.

There is one skill per project badge, plus a slam-status skill that checks all six. Each skill audits a repository against the badge requirements, reports gaps with evidence, drafts only what the maintainers confirm, and ends with a submission checklist.

Quick start

Install the skills, then open your agent in the repository you want to check and ask: "Where does this repo stand in the Security Slam?"

Claude Code:

claude plugin marketplace add security-slam/skills
claude plugin install security-slam-skills@security-slam
claude "/security-slam-skills:slam-status"

Any other agent:

npx skills add security-slam/skills

That question runs slam-status, which is read-only. It reports where you stand on all six badges and suggests one next step.

Status

Early preview. The Spring 2026 versions of the skills ran end to end on four repositories, and each one earned evidence for all six badges using the skills. The Fall 2026 badge pages moved Mechanizer and Defender evidence to grc.store, and no repository has been through that flow with the skills yet:

Repository	What it is	Baseline scan	grc.store target
security-slam/skills	This repository: Markdown and YAML, released on each plugin change	Scanner action	Pending first run: publish-results.yaml publishes to eddie-knight/skills, a personal namespace bound to this repository
security-slam/website	A TypeScript site with npm dependencies, deployed and released on every merge, with existing Gemara catalogs	Scanner action	Not yet
privateerproj/privateer	A Go CLI that ships release binaries with GoReleaser	Scanner action	Not yet
privateerproj/privateer-sdk	A Go library that inherits its security policy and contributing guide from an org .github repository	Scanner action	Not yet

The Spring bestpractices.dev Baseline Level 1 entries (skills, website, privateer, privateer-sdk) stay up but no longer count toward Defender.

Tested with Claude Code (plugin marketplace) and Codex (installed with npx skills). On privateer-sdk, Codex picked slam-status from a plain question and produced the same report as Claude Code.

Not yet tested:

• The grc.store publish workflow (mechanizer option 2) and the defender flow that reads the published result. Both were written against a working run on revanite-io/pvtr-aws-s3, not against a run the skills produced.
• The GV and VM documentation controls chronicler now covers, and drafting from the OSPS Templates.
• Projects that publish container images or registry packages.
• OSPS Baseline Level 2 and 3 paths, such as release signing, SBOMs (software bills of materials), and release verification docs. All four repositories stopped at Level 1.
• The prose self-assessment option of inspector, and its hand-off to the gemara-ai plugin.
• Projects hosted outside GitHub.
• Agents other than Claude Code and Codex, and every skill except slam-status outside Claude Code.

The skills pin the specification versions listed under Versions Targeted. When those specifications change, the skills can lag behind until a release updates them.

If a skill gets something wrong on your project, please report it.

Install

Prerequisites

Every skill starts by running the OSPS Baseline scanner locally, so the report starts from real results instead of inference. That needs three tools on your PATH:

• gh, signed in. The scanner reads the repository with your token, read-only.
• pvtr, the Privateer CLI: brew install privateerproj/tap/pvtr, or the install script in its README. The skills install and update the openssf/github-repo scanner plugin themselves.
• yq to read the results.

The cleaner skill also needs cue to validate the Security Insights file.

Claude Code

From inside Claude Code:

/plugin marketplace add security-slam/skills
/plugin install security-slam-skills@security-slam

Or from your shell:

claude plugins marketplace add security-slam/skills
claude plugins install security-slam-skills@security-slam

To try a local checkout, pass the directory path in place of security-slam/skills.

Other agents

skills installs Agent Skills into most compatible agents:

npx skills add security-slam/skills

It asks which agents to install into. Add --agent <name> to pick one, --skill <name> to install a single skill, or --global to install for your user instead of the current project. The installer sends anonymous usage data by default. Set DO_NOT_TRACK=1 to turn that off.

You can also copy the folders under security-slam-skills/skills/ into your agent's skills directory.

Skills

Not sure where a project stands? Run slam-status first. Otherwise start with Cleaner. Every other badge adds links to the Security Insights file it creates.

Skill	Badge requirement
cleaner	Complete and validate the project's Security Insights YAML file.
chronicler	Complete every documentation-shaped OSPS Baseline control (DO, GV, and VM) for the project's maturity level, drafting policies from the OpenSSF OSPS Templates.
inspector	Complete a Gemara threat assessment or an OSPS self-assessment.
mechanizer	Wire a live, recurring OSPS Baseline scan into the default branch, as the scanner action or the grc.store publish workflow, and record it in Security Insights.
defender	Reach a passing OSPS Baseline result for the project's maturity level, published on grc.store, with Security Insights evidence for the controls the scanner cannot check.
cra	Voluntary EU Cyber Resilience Act (CRA) readiness. Details for this badge will be released on October 12, 2026, and the skill says so until then.
slam-status	Check a repo against all six badges at once and recommend which one to work on next. Read-only.

Every skill begins with a local run of the OSPS Baseline scanner over each repository in the project, and infers only what the scanner cannot check. Your agent picks a skill when your request matches its description. You can also ask for one by name, for example "use the cleaner skill". In Claude Code, /cleaner works too.

The individual recognitions (Security Advocate, Security Champion, Advisor) go to people, not projects, so they have no skills.

Versions Targeted

Dependency	Version
OSPS Baseline	2026-08-28
Security Insights schema	2.2.0
Gemara schema	1.5.0
grc.store publish workflow (revanite-io/pvtr-publish-results)	v1
OSPS Baseline scanner plugin (openssf/github-repo on grc.store)	latest at run time; 0.31.0 tested

The inspector skill hands Gemara authoring to the gemara-ai plugin when it is installed, and explains how to migrate catalogs written for Gemara 0.19.x.

Update

Claude Code:

claude plugins marketplace update security-slam
claude plugins update security-slam-skills@security-slam

Other agents:

npx skills update

Releasing

Claude Code caches plugins by version, so Claude Code users only receive changes that come with a version bump. npx skills installs from the repository, so its users get the latest main.

• Any PR that changes security-slam-skills/ or .claude-plugin/marketplace.json must bump version in security-slam-skills/.claude-plugin/plugin.json and metadata.version in .claude-plugin/marketplace.json to the same value. The version-check workflow enforces this.
• The auto-labeler adds the release label to those PRs, and merging one publishes a release automatically.
• CI, docs, and other repo changes never trigger a release.

Uninstall

Claude Code:

claude plugins uninstall security-slam-skills@security-slam
claude plugins marketplace remove security-slam

Other agents:

npx skills remove cleaner chronicler inspector mechanizer defender cra slam-status

Security

See SECURITY.md to report a vulnerability. The project voluntarily documents its practices in CRA-READINESS.md.

License

Apache 2.0