context-manager
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch external documentation and process user-provided summary files, creating a surface where malicious instructions hidden in external data could influence agent behavior.\n
- Ingestion points: External documentation via 'FETCH' and local summaries via 'HARVEST' (e.g., ANALYSIS.md).\n
- Boundary markers: None identified; no instructions are provided to isolate embedded commands in external data.\n
- Capability inventory: The skill uses a bash script ('router.sh') as a gateway, representing potential shell access, though current code only echoes status.\n
- Sanitization: No evidence of validation or filtering for ingested content.\n- [METADATA_POISONING]: A discrepancy exists between the author identified in the skill metadata ('opencode') and the provided author context ('darrenhinde'). While not explicitly malicious, metadata inconsistencies can be deceptive.
Audit Metadata