component-gallery
Fail
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: CRITICALCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: Automated reputation scanners flagged
references/index.jsonandreferences/design-system-index.mdas malicious (FileRepMalware). This suggests the inclusion of files that have been pre-identified as suspicious by security software. - [COMMAND_EXECUTION]: The skill invokes local command-line tools
firecrawlandrlamaviasubprocess.runinscripts/ingest.pyto manage data ingestion and RAG database operations. - [COMMAND_EXECUTION]: In
scripts/query.py, the skill executes an external Python script (rlama_retrieve.py) located in a different skill's directory (~/.claude/skills/rlama/) usingsubprocess.run. - [EXTERNAL_DOWNLOADS]:
scripts/ingest.pydownloads content from an unverified third-party GitHub repository (inbn/component-gallery), whilescripts/fetch_mesh3d.pyfetches directory listings frommesh3d.gallery. - [INDIRECT_PROMPT_INJECTION]: The skill's RAG system ingests untrusted markdown data from the web and third-party repositories. This content is stored and later provided to the agent, creating a vector for indirect prompt injection as there are no visible sanitization steps or boundary markers.
- Ingestion points:
scripts/ingest.py(GitHub fetch and crawl) andscripts/fetch_mesh3d.py(web fetch). - Capability inventory: Shell command execution via
subprocess.runand network communication viaurllib.request. - Sanitization: No sanitization or escaping of the ingested markdown content is present in the provided scripts.
- [DYNAMIC_EXECUTION]: The
scripts/query.pyscript dynamically builds and executes a command that invokes a script from an external skill path, which introduces a dependency on the integrity of code not directly managed by this skill.
Recommendations
- CRITICAL: 2 file(s) identified as malware by FileRep - DO NOT USE
Audit Metadata