prd-to-kandev-tasks
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted product requirements (PRDs, specs, roadmap items) and uses this content to generate task descriptions. These descriptions are explicitly intended to serve as the system instructions or 'launch prompts' for other agents.
- Ingestion points: Processes PRD text, file paths, or task descriptions provided as input to decompose work in
SKILL.md(Step 2). - Boundary markers: No specific delimiters or 'ignore embedded instructions' wrappers are defined for the interpolated PRD content within the generated task briefs in
references/task-blueprint.md. - Capability inventory: The skill possesses the ability to read local project files for discovery and create new tasks with rich descriptions via MCP (
create_task_kandev) or the CLI (kandev task create), which can launch additional autonomous agents as noted inSKILL.md(Step 4 and Step 5). - Sanitization: While the skill enforces 'quality gates' to ensure tasks are descriptive and actionable, it lacks specific instructions to sanitize or escape potentially malicious instructions embedded within the source PRD text before it is passed to downstream agents.
Audit Metadata