publish-agent-skill
Pass
Audited by Gen Agent Trust Hub on Oct 2, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using CLI tools such as apm, claude, and npx skills to manage skill installations and marketplaces.
- [EXTERNAL_DOWNLOADS]: The workflow involves fetching skill content and manifests from GitHub repositories and the APM registry, which is the intended function for a publication utility.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external agent skill directories containing natural language instructions. 1. Ingestion points: Skill files like SKILL.md are read by the verification script. 2. Boundary markers: None; the tool relies on binary integrity comparison. 3. Capability inventory: Uses apm, claude, and npx for installation and node for file comparison. 4. Sanitization: None; the process ensures file identity rather than content safety.
- [DYNAMIC_EXECUTION]: The test file tests/verify-install.test.mjs uses spawnSync to execute a verification script, which is a standard pattern for testing CLI-based tools.
Audit Metadata