trailmark-summary
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface through tool output poisoning because it processes untrusted codebase metadata and returns the output to the agent context.
- Ingestion points: In
SKILL.md(Step 2 and Step 3), the skill analyzes an external codebase directory provided via the{args}parameter to extract languages, entry points, and dependencies. - Boundary markers: Absent. The output from the
trailmarkexecution is not encapsulated within clear delimiters or guardrails to prevent the agent from interpreting text inside the codebase as instructions. - Capability inventory:
SKILL.mdexecutes local shell commands and an inline Python script via theBashtool. - Sanitization: Absent. The resulting list of detected languages and structural overview text is forwarded to the agent context without escaping or filtering.
Audit Metadata