tighten-file
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is entirely instructional and handles text-based refinement tasks. It does not perform network operations, access sensitive credentials, or execute arbitrary system commands.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided instruction files, which serves as a potential vector for indirect prompt injection.\n
- Ingestion points: Instruction files provided for tightening in SKILL.md.\n
- Boundary markers: None explicitly mentioned in the workflow, though Step 1 (Pinning meaning) acts as a logical check for intended behavior.\n
- Capability inventory: The skill is restricted to text analysis and invoking an internal refinement tool; no dangerous system capabilities are present.\n
- Sanitization: The skill does not implement technical sanitization, relying on the agent's proof-reading step to catch deviations from original intent.
Audit Metadata