skill-context-detection
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill detects context by scanning user prompts for specific keyword indicators (e.g., 'market', 'ROI', 'API', 'build'). A crafted prompt containing keywords from a different context could lead to context confusion, causing the agent to select sub-agents or output formats that do not align with the user's actual intent.
- Ingestion points: User prompt content is analyzed for keyword scoring as described in the 'Detection Algorithm' section of SKILL.md.
- Boundary markers: None identified; the detection logic scans the entire prompt for signals.
- Capability inventory: The detected context determines sub-agent selection (e.g., selecting 'security-auditor' for dev context vs 'exec-communicator' for knowledge context) and influences how prompts are framed for external CLI tools.
- Sanitization: No input sanitization or filtering is performed on the prompt content before keyword analysis.
- [DATA_EXPOSURE]: The skill accesses the local file system to read and write configuration and preference files located at '
/.claude-octopus/config/knowledge-mode' and '/.claude-octopus/preferences.json'. These operations are restricted to namespaced directories and are used for legitimate configuration management.
Audit Metadata