iterate-revision

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes content from manuscript files and outputs from writer/reviewer subagents, interpolating them into instructions for further agent actions. This creates a surface for indirect prompt injection where malicious content in a processed file could override agent behavior.
  • Ingestion points: Manuscript section files (.tex, .md, .txt), paper.json fields (hypothesis, title), and the generated outputs from the @paper-writer and @reviewer subagents.
  • Boundary markers: The skill uses simple textual labels like 'Current section content:' to separate data from instructions but lacks robust delimiters or explicit instructions to the agent to treat interpolated content as untrusted data.
  • Capability inventory: The skill is configured to write to and edit specified project files and dispatch subagents through orchestration primitives.
  • Sanitization: While structured output from the reviewer is validated against a schema, the writer's prose and the initial manuscript content are handled as raw strings and interpolated into prompts without escaping or sanitization.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 09:21 AM
Security Audit — agent-trust-hub — iterate-revision