reclaim
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill incorporates strong safety boundaries, explicitly forbidding the modification of the filesystem, such as deletion, pruning, or configuration changes, and strictly prohibits the extraction of sensitive data like credentials, private keys, and personal documents.
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes local system metadata and project manifests, which represents a potential attack surface for indirect prompt injection if files or manifests contain malicious instructions.
- Ingestion points: System-wide file metadata, project manifests (e.g., package.json, lockfiles), and directory structures referenced in SKILL.md and its associated reference files.
- Boundary markers: The skill explicitly instructs the agent to treat these sources as data for measurement and to avoid reading file contents for most categories, specifically avoiding secrets and private records.
- Capability inventory: The skill utilizes local, read-only inspection tools such as 'du', 'df', and 'docker' to collect storage metrics.
- Sanitization: The risk is mitigated by the metadata-centric logic and explicit 'read metadata before file contents' rules.
Audit Metadata