policy-algebra

Pass

Audited by Gen Agent Trust Hub on Apr 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill operates locally and follows a predictable execution flow. It does not perform network operations or access sensitive credentials.\n- [COMMAND_EXECUTION]: The skill executes its internal logic via a TypeScript CLI tool to perform rule verification tasks.\n
  • Evidence: pnpm exec tsx .claude/skills/policy-algebra/lib/cli.ts verify is used for drift detection and canonicalization.\n
  • Context: This is an intended functionality for processing governance rules within the local environment.\n- [PROMPT_INJECTION]: The skill processes untrusted input to generate rules, creating a surface for indirect prompt injection.\n
  • Ingestion points: User-provided intent strings and local file contents are used as input for drafting governance rules.\n
  • Boundary markers: Data is interpolated into the /grill-me skill's instructions within quotes to separate user intent from the agent's procedural instructions.\n
  • Capability inventory: The skill has the ability to read and write files and invoke other local skills like /grill-me and /review-loop.\n
  • Sanitization: No explicit filtering or sanitization of input data is performed, relying instead on the agent's internal safety filters and the review loop process.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 13, 2026, 04:27 PM