dispatching-parallel-agents
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists of markdown documentation outlining architectural patterns for multi-agent delegation. It does not contain code, scripts, or instructions that deviate from its stated purpose.\n- [EXTERNAL_DOWNLOADS]: The documentation references external skills from the author's repository (
krzysztofsurdy/code-virtuoso). These are listed for integration and workflow purposes and follow the project's standard structure.\n- [PROMPT_INJECTION]: The instructions do not contain attempts to bypass safety filters or override system constraints. Instead, the skill emphasizes 'clean context' and strict boundaries for subagents to ensure predictable behavior.\n- [COMMAND_EXECUTION]: The skill suggests using tools likeBashfor git-based filesystem isolation (e.g.,git worktree). These are recommended as safety mechanisms to prevent overlapping writes and do not involve arbitrary or dangerous execution patterns.\n- [INDIRECT_PROMPT_INJECTION]: The orchestration patterns include validation and synthesis phases designed to mitigate risks from processing subagent outputs.\n - Ingestion points: The orchestrator receives results from subagents (defined in
references/isolation-and-merging.md).\n - Boundary markers: The skill mandates clean context, self-contained briefs, and explicit out-of-scope boundaries for all subagents.\n
- Capability inventory: Uses
Read,Grep,Glob, andBashfor orchestration and synthesis tasks.\n - Sanitization: Employs mandatory validation against briefs and 'synthesis over concatenation' to prevent accidental obedience to instructions embedded in subagent results.
Audit Metadata