pathfind
Pass
Audited by Gen Agent Trust Hub on Aug 16, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external issue trackers and documentation, which could contain malicious instructions. The skill proactively addresses this by instructing the agent to treat external content as passive data and ignore embedded instructions.
- Ingestion points: The skill reads map tickets, brownfield scans, and issue trackers (e.g., docs/agents/issue-tracker.md).
- Boundary markers: No technical delimiters are specified in the instructions, but the skill defines a logical boundary by treating external data as passive.
- Capability inventory: The skill uses sub-skills like clarify-decisions, research, and run-spike for prototyping.
- Sanitization: The instructions explicitly tell the agent to ignore strings like 'IGNORE PRIOR RULES' found in issue bodies.
- [PROMPT_INJECTION]: Static detectors flagged instructions related to ignoring prior rules; however, these are defensive rationalizations provided to help the agent resist injection attempts rather than malicious overrides.
Audit Metadata