parallel-orchestration
Pass
Audited by Gen Agent Trust Hub on Nov 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to orchestrate tasks based on data ingested from external sources (GitHub Issues, Specs, and ADRs), which creates a surface for indirect prompt injection where malicious instructions could be embedded in those sources.
- Ingestion points: Data is ingested during the initial task decomposition phase (described in SKILL.md, Flow step 1) and during the authoring of execution instructions (detailed in references/requirement-continuity.md).
- Boundary markers: The framework utilizes immutable snapshots, exact commit SHAs, and fencing identities (execution_generation) to pin task context and prevent unauthorized state changes.
- Capability inventory: The orchestration layer spawns workers with defined 'allowed_tools', 'filesystem_policy', and 'network_policy', potentially allowing for tool execution, file system access, and network operations depending on the assigned profile.
- Sanitization: The 'requirement-continuity.md' file defines explicit procedures for 'Canonical requirement continuity', requiring agents to verify instructions against project-local canonical sources and prohibiting the adoption of unauthorized instructions injected by upstream assistants.
Audit Metadata