lean-run

Pass

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/check.py script allows the agent to execute arbitrary shell commands using subprocess.Popen. This is a core capability used for running tests, scripts, and builds as described in the SKILL.md instructions. The agent is directed to use this wrapper to capture logs and manage process execution.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources such as source code, logs, and pull request descriptions. This represents an attack surface for indirect prompt injection. The skill mitigates this by providing extensive documentation in references/parser-integrity.md and instructions in SKILL.md (Step 5: "Own inputs") that require the agent to sanitize inputs by converting them to plain built-in types and validating them against strict schemas before use.
  • Ingestion points: scripts/view.py (reading file content) and scripts/check.py (capturing command output).
  • Boundary markers: The instructions emphasize identifying format rules and byte ranges to establish authoritative records, rather than relying on layout.
  • Capability inventory: Execution of shell commands via scripts/check.py and file reads via scripts/view.py.
  • Sanitization: The skill mandates converting caller-supplied values into owned plain types and provides a detailed strategy for parser integrity.
  • [DATA_EXFILTRATION]: The agent has the ability to read local files via scripts/view.py and potentially access sensitive environment variables or files using scripts/check.py. While the skill does not explicitly command sending data externally, the capability to read and execute commands constitutes an exposure surface. The skill manages this risk through a classification system (Step 4) that flags security-sensitive work as HIGH risk and requires independent review and verification.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 10, 2026, 02:14 PM