runbook-writer
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The instructions focus on coordination and quality control without any patterns that would attempt to bypass safety filters or override core agent behavior.
- [DATA_EXFILTRATION]: While the skill accesses sensitive technical documentation (HLD, LLD, etc.) to generate its output, there are no network operations or commands directed at exfiltrating this data to external or untrusted domains.
- [OBFUSCATION]: Analysis of all files, including the SVG asset, revealed no hidden commands, encoded strings, or obfuscated URLs.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design documents. It manages this risk through a structured 'Context Isolation' approach and a two-stage review process (Spec Compliance and Quality Review).
- Ingestion points: PRD, HLD, LLD, API Contract, and Guardrails files referenced in Phase 0.
- Boundary markers: Context is structured using markdown templates with defined sections to isolate data provided to subagents.
- Capability inventory: Subagent dispatch via Task tool, user interaction via AskUserQuestion, and local file writing.
- Sanitization: Relies on extraction of specific constraints and a two-stage review process involving specialized subagents.
- [COMMAND_EXECUTION]: Example commands in output templates (e.g., kubectl, curl) are provided as documentation placeholders for the end-user and are not intended for execution by the agent itself.
- [DYNAMIC_CONTEXT_INJECTION]: No use of the dynamic execution syntax (!command) was detected in the skill instructions.
Audit Metadata