akf-trust-metadata
Warn
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to install packages from public registries without version pinning. These packages are not from verified or trusted organizations.\n
- Evidence: 'pip install akf' and 'npm install akf-format' in SKILL.md.\n- [INDIRECT_PROMPT_INJECTION]: The skill reads and inspects metadata from external files, which creates a surface for indirect prompt injection if the files are attacker-controlled.\n
- Ingestion points: 'akf read ' and 'akf inspect ' commands documented in SKILL.md.\n
- Boundary markers: Absent. The skill lacks instructions for the agent to use delimiters or ignore embedded instructions within metadata.\n
- Capability inventory: 'akf stamp' (file modification), 'pip install', 'npm install' (command execution).\n
- Sanitization: Absent. The agent is directed to use metadata for auditing and compliance tasks without validation steps.
Audit Metadata