external-pattern-integrator
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's core function is to ingest instructions from external sources (URLs or local files) and integrate them into local skill and agent definitions. This creates a significant surface for indirect prompt injection, as external content is treated as a source of "best practices" and "patterns" for local instructions.
- Ingestion points: External files and URLs provided via $ARGUMENTS and fetched in Step 1.2.
- Boundary markers: No specific boundary markers or "ignore instructions" warnings are implemented when processing external content.
- Capability inventory: The skill can read local files, fetch remote content via curl/WebFetch, write to local .md files in .claude/ and plugins/, and execute shell commands via uv.
- Sanitization: There is no evidence of sanitization or filtering of the external content before it is used to generate or modify local instructions.- [EXTERNAL_DOWNLOADS]: The skill uses tools like curl or WebFetch to download content from arbitrary, user-supplied URLs to /tmp/ for analysis.- [COMMAND_EXECUTION]: The skill executes the command
uv run prek run --files {all modified files}in Phase 3.1. While this appears to be a linting step, it involves running a subprocess with arguments derived from modified files.
Audit Metadata