comet-native
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external source materials such as files, attachments, or links provided by the user. This creates a surface for indirect prompt injection where instructions embedded in these materials could attempt to subvert the agent's behavior.\n
- Ingestion points: User-provided files, attachments, and links used as requirements sources (identified in
SKILL.mdunder the 'Shape' section).\n - Boundary markers: The skill explicitly instructs the agent that instructions found within source materials must be treated as content only and cannot override the user's request or project rules.\n
- Capability inventory: The skill executes shell commands via the
cometCLI and writes to project files likebrief.mdandspecs/.\n - Sanitization: Relies on the agent's ability to distinguish between data content and instructions as directed by the skill.\n- [COMMAND_EXECUTION]: The skill performs shell command execution using the
cometCLI (e.g.,comet native,comet task) to manage the workflow. It passes user-provided input, such as the original request, directly into command arguments.
Audit Metadata