skill-curator

Pass

Audited by Gen Agent Trust Hub on Jul 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as its core function is to ingest and process untrusted external source material to generate new agent instructions.
  • Ingestion points: Untrusted data enters the context through the research dossier workflow defined in SKILL.md and references/research-dossier.md.
  • Boundary markers: The skill provides "Paraphrase boundary" and "Paraphrase audit" instructions to mitigate the reproduction of distinctive source phrasing.
  • Capability inventory: The skill possesses capabilities to create and modify files and directories, and to execute shell commands and local scripts such as grep and just check.
  • Sanitization: A required paraphrase audit step is used to rewrite or remove distinctive phrasing before a generated skill is handed off for review.
  • [COMMAND_EXECUTION]: The skill's workflow and evaluation scripts involve executing several local commands.
  • Workflow commands: Instructions in SKILL.md direct the agent to execute grep, just check, and scripts/check-skill-static.py.
  • Script execution: The evals/run-static-checks.sh script performs file checks and uses grep for pattern matching. It also dynamically sources a local library file, scripts/static-check-lib.sh, by searching the project's directory hierarchy.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 12, 2026, 07:07 PM
Security Audit — agent-trust-hub — skill-curator