product-tracking-instrument-new-feature
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied feature specifications and existing repository configuration files to generate tracking updates. This creates an ingestion surface where malicious instructions hidden in feature documentation could attempt to influence the agent's logic during the tracking plan update process.
- Ingestion points: The agent reads existing telemetry files (
.telemetry/tracking-plan.yaml,.telemetry/product.md) and feature descriptions provided in the conversation context. - Boundary markers: The instructions include guidelines for the agent to read the plan fully and reason silently, which provides some context isolation, though explicit data delimiters are not enforced.
- Capability inventory: The skill is restricted to reading and writing Markdown and YAML documentation within the
.telemetry/directory. No network operations or shell commands are performed. - Sanitization: The output is strictly formatted according to provided YAML and Markdown templates, which limits the impact of potential injection to the content of the documentation files.
Audit Metadata