gathering-security

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests user-provided security specifications and codebase files to be processed by autonomous subagents.
  • Ingestion points: Security specifications in the SUMMON phase and existing codebase files across the WEAVE, AUDIT, and HARDEN phases.
  • Boundary markers: The conductor uses structured markdown headers (e.g., ## Your Mission, ## Your Input) in subagent dispatch templates to delimit instructions from processed data.
  • Capability inventory: Dispatched subagents have access to file search (gf), git operations (gw), and file modification capabilities.
  • Sanitization: No explicit content sanitization is performed on input files; the skill relies on model isolation (adversarial fresh eyes) and automated CI gate checks to verify outputs.
  • [COMMAND_EXECUTION]: The skill executes local command-line tools to verify the integrity of the code following modifications.
  • Evidence: The skill uses gw dev ci --affected --fail-fast and pnpm install as mandatory gate checks in Phase 2, 3, 4, and 6 to ensure compilation and functional safety.
  • [DYNAMIC_EXECUTION]: The skill dynamically manages the execution context by dispatching specialized subagents to perform distinct security tasks.
  • Evidence: The conductor orchestrates complex handoffs between roles (e.g., Spider to Raccoon, Turtle to Spider for iterations), passing structured data and instructions to subagents at runtime based on the security scope.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 01:31 AM