This bundle is read-mostly, dependency-free, and offline. It reads your source to discover ground truth and writes only Storybook artifacts. The statements below are what the skills.sh Gen Agent Trust Hub content-safety review looks for, and they are mechanically verifiable from the shipped files.
- The bundle ships no
package.json, no lockfile, no vendorednode_modules. Skills are SKILL.md + portablebash/tsx. There is no install-time code execution. - Real Socket result:
socket manifest autoreports "unable to discover any targets for which we can generate manifest files" — there is no dependency manifest for a supply-chain engine to flag. (Seedocs/pre-publish-audit.mdfor the command.) - The wrapper components (
shared/wrappers/*.tsx) are copied into the consuming project, where they use only that project's already-present React/Storybook. The one optional peer dep (@react-three/fiber, forR3FCanvas) is only suggested in a printed reminder — never auto-installed.
- No shipped script or wrapper makes a network call. There is no
curl/wget, no/dev/tcp, nofetchto a remote host, no telemetry, no analytics. Verify:grep -rnE 'curl|wget|/dev/tcp|fetch\(|http://|https://' shared/ skills/returns only documentation URLs (skills.sh, github.com, storybook.js.org), never a runtime call. npx storybook ai setup(invoked bysb-setup) is the consuming project's own Storybook installer, run explicitly by the user — not a network call this bundle makes.
- No script reads environment secrets,
~/.aws,~/.ssh,.env, or 1Password. There are no API keys, tokens, or credentials anywhere in the bundle. - Discovery scripts read source code and CSS only (
src/**,*.css,tokens.json,package.jsonfor framework detection by name, never its values).
Every shipped script writes only to:
.storybook/— discovery JSON (project-inventory.json,flows.json,component-states.json,prop-shapes.json,design-system-health.json),wrappers/, the audit ledger (audit/{findings,plan,status}.md), and the Storybook config (main.ts,preview.*).src/stories/**andsrc/explore/**— generated*.stories.tsx.
Application code under src/components/** is never modified by discovery, build,
or validation skills. The single skill that touches app code is sb-ship, and it
copies a graduated experiment (cp, never git mv), preserving the original as
history. Writes are atomic (tempfile → os.replace) so an interrupted run never
leaves a half-written JSON.
- No
rm -rf, nogit push, nogit reset --hard, no force operations. sb-shipadds files; it does not delete or rewrite history.
Two skills carry a MEDIUM scanner WARN. Both are heuristic false positives that fire on code execution (which every functional skill does), not real vulnerabilities. Provenance:
sb-setup— Snyk W012 "unverifiable external dependency / runtime URL (0.90)". This is thenpx storybook ai setupline.storybookis the official Storybook npm package (@storybook/cli), not a URL and not bundled or controlled by this skill — it is the project's own Storybook installer, run explicitly by the user. The skill ships zero runtime dependencies and makes no network calls of its own (see Network — none above). Removing the step would defeat the skill's sole purpose (deferring to Storybook's official onboarding instead of reinventing it).sb-wrappers— Socket "Security · SKILL.md · MEDIUM". This is thescaffold-wrapper.shinvocation. That script is a local bash file vendored inside the skill — it only copies the bundled.tsxwrapper components into the project's.storybook/. No network, no install, no external/remote code.
Both are run explicitly by the user/agent, never silently, and both are reviewable in full from the shipped source. We treat these WARNs as expected for a code-running skill; this section is the appeal evidence for a re-audit.
This bundle lives in strongeron/agent-skills. Report any security concern via a
GitHub issue on that repo. Because the bundle is offline and dependency-free, the
realistic surface is limited to the correctness of the bash/tsx it ships — reviewable
in full from source.