ship
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill documents various deployment commands using the
pnpmpackage manager, such asdeploy:plan,deploy:prod, anddeploy:rollback. It also references the use ofVarlock, which supports shell command execution via anexec()function within local environment files to resolve secrets safely.\n- [INDIRECT_PROMPT_INJECTION]: The skill incorporates data ingestion points that could theoretically be used for indirect prompt injection, although it includes documentation on sanitization.\n - Ingestion points: Ingestion of
sitemap.xmlfor post-deploy verification and Cloudflare Worker crash logs from R2 buckets as diagnostic data.\n - Boundary markers: Not explicitly defined in the instructional text for the external data sources.\n
- Capability inventory: Command execution through
pnpmscripts, file system writes to local recovery directories (.rat/deploy/), and interactions with Cloudflare's API.\n - Sanitization: The documentation explicitly calls for the redaction of credentials, private identifiers, and sensitive request/response metadata within error reports and crash archives.
Audit Metadata