run-history-skill-builder

Pass

Audited by Gen Agent Trust Hub on Jul 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes a local validation script scripts/validate_skill_package.py which uses the py_compile module to perform syntax checks on generated Python code.
  • [DATA_EXFILTRATION]: The skill has an extensive data ingestion surface, including reading workspace files, logs, and artifacts. This capability allows access to sensitive information, although instructions specifically mandate the exclusion of credentials and personal data.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted inputs from logs and transcripts that could contain hidden instructions aimed at influencing the generated skill's behavior. 1. Ingestion points: SKILL.md (logs, workspace files), references/history-mining.md. 2. Boundary markers: Present in workflow instructions (locking scope and output). 3. Capability inventory: File writing and execution of scripts/validate_skill_package.py. 4. Sanitization: Instructions to strip credentials and private paths are provided in references/skill-design-protocol.md.
  • [EXTERNAL_DOWNLOADS]: The instructions in references/open-source-pattern-mining.md encourage cloning and downloading external repositories for pattern analysis, which introduces a potential vector for ingesting malicious configurations or scripts from unverified sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 9, 2026, 05:49 AM
Security Audit — agent-trust-hub — run-history-skill-builder