eng-spec
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and interpolates it into prompts for multiple sub-agents, creating a vulnerability to malicious instructions embedded in the source documents.\n
- Ingestion points: PRD files from the project-specified
PM_OS_DIRand UX/UI design assets (screens, wireframes, flows) fromUXD_OS_DIRare loaded into the agent's context.\n - Boundary markers: The review prompts in
PANEL-PROMPTS.mduse simple bracketed delimiters (e.g.,[PRD content]) to separate data from instructions. There are no explicit instructions for the sub-agents to ignore potentially malicious directions within the interpolated content.\n - Capability inventory: The skill writes engineering artifacts (TDD, SPEC, ADRs) to the local filesystem, updates a pipeline tracking file (
PRD-PIPELINE.md), and executes thehq.sync-contextcommand.\n - Sanitization: No evidence was found of input sanitization, filtering, or escaping of the content from the PRD or UX assets before they are passed to the sub-agents.\n- [COMMAND_EXECUTION]: The skill workflow concludes with the execution of the
hq.sync-contextcommand in Step 10. While this appears to be an intended synchronization step for the engineering context, it is an automated shell command execution triggered by the skill's logic.
Audit Metadata