context-management
Pass
Audited by Gen Agent Trust Hub on May 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns detected. The skill consists of instructional text for context optimization, state management, and handoff procedures.
- [DATA_EXPOSURE]: The skill outlines processes for reading files and logs via the 'Read' and 'Grep' tools. These operations are limited to the agent's authorized workspace and are used for the intended purpose of context building.
- [INDIRECT_PROMPT_INJECTION]: The skill involves processing untrusted external data such as error logs and third-party documentation. While this presents a potential surface for indirect prompt injection, the skill's emphasis on 'distillation' (converting raw logs to 2-line summaries) and 'collapsing facts' serves as a manual sanitization process that mitigates the risk of the agent executing embedded instructions.
- Ingestion points: Files and logs read via 'Read' and 'Grep' tools.
- Boundary markers: None explicitly defined in instructions.
- Capability inventory: File reading (Read), searching (Grep).
- Sanitization: None specified beyond natural language summarization.
Audit Metadata