workflow-research

Warn

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill instructs the agent to execute local JavaScript files using the node runtime (e.g., .claude/scripts/lib/workflow-skill-contract.cjs and .claude/scripts/lib/workflow-manifest.cjs). This allows the execution of arbitrary logic contained within the project's subdirectory.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to research and synthesize information from external web sources, which represents a surface for indirect prompt injection. * Ingestion points: External web content fetched via the $web-research child skill. * Boundary markers: The workflow utilizes a mandatory $knowledge-review quality gate to verify artifacts and check for bias, which provides a layer of oversight. * Capability inventory: The skill has permissions to spawn sub-agents, write research artifacts to the local filesystem (tmp/reports/, docs/knowledge/), and execute local scripts via shell commands. * Sanitization: The instructions do not specify programmatic sanitization or escaping of the raw external data before it is processed by the model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 10, 2026, 08:16 PM