Built-in Skills

Use built-in skills by name. No local skill file, build step, schema, or SDK is required.

security-review

security-review is Warden’s baseline application security pass. It looks for real exploitable vulnerabilities in changed code.

warden add security-review
warden --skill security-review

code-review

code-review looks for correctness bugs and behavioral regressions in changed code.

warden add code-review
warden --skill code-review

integrity-review

integrity-review looks for changes that would harm the organization through their own behavior, or that hide what they do: destructive operations, date or condition triggers, install and build hooks, dependency and registry tampering, disabled tests or security tooling, audit logging removal, infrastructure privilege grants, obfuscated code, misleading comments, text aimed at automated reviewers, and changes outside the pull request’s stated scope.

warden add integrity-review
warden --skill integrity-review

It complements the other two: security-review finds weaknesses an outside party can exploit, code-review finds unintended bugs, and integrity-review finds changes where the code itself is the problem. Shared cases such as a concealed auth bypass are reported once by each skill from its own angle.

Do not narrow integrity-review with paths. Lockfiles, vendored code, generated files, fixtures, CI config, and infrastructure are where harmful changes hide because reviewers skip them.

Most hunks never reach a model. Warden first checks each hunk in code for install or prepare hooks, encoded or high-entropy strings, deletion-heavy hunks, workflow and CODEOWNERS changes, date or random checks next to deletes, destructive commands, removed or suppressed controls, privilege grants in infrastructure files, and text addressed to automated reviewers. Hunks that match none are skipped. Text addressed to reviewers in the PR title or body sends every hunk through.

Run it when a PR opens, when it leaves draft, and on each push. On a push it reviews only the hunks that push touched; when GitHub cannot compare the push, such as after a force-push, it reviews the whole PR diff.

[[skills]]
name = "integrity-review"

[[skills.triggers]]
type = "pull_request"
actions = ["opened", "ready_for_review", "synchronize"]
draft = false

Add ready_for_review to the workflow’s pull_request types as well.

How They Fit Together

Skill	Finds	Does not find
`security-review`	Exploitable weaknesses: auth bypass, injection, XSS, SSRF, secrets exposure	Changes whose own behavior is the harm
`code-review`	Unintended correctness bugs and regressions	Security or integrity issues
`integrity-review`	Destructive, concealed, or trigger-gated changes; reviewer-directed text	Exploitable weaknesses, ordinary bugs

Built-ins are first passes. Add project-specific skills for domain rules, framework conventions, API contracts, and product-specific invariants.