hrt-adversarial-authoring

Pass

Audited by Gen Agent Trust Hub on Sep 11, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading external markdown files and using their content to guide subagent authoring and review tasks, creating a surface for indirect instructions to influence agent behavior.
  • Ingestion points: External data enters the context through grounding_inputs (specifically discovery.md, proposal.md, and specs/**/*.md) and an optional dark.md file.
  • Boundary markers: The skill employs mitigation by spawning review subagents in "fresh isolated contexts" with no memory of the author's internal reasoning, which helps prevent logic contamination.
  • Capability inventory: The orchestrator has the capability to spawn new agent sessions, write finalized artifacts to the file system, and perform automated text resolutions ("Auto-resolve") based on reviewer feedback.
  • Sanitization: All merged findings and notes are passed through a humanizer skill before being presented to the user or used for resolution, providing a layer of transformation that can help neutralize simple embedded instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 11, 2026, 02:24 AM
Security Audit — agent-trust-hub — hrt-adversarial-authoring