grace-multiagent-execute

Warn

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is designed to parse and execute shell commands extracted from documentation files. Step 3 and Step 6 instructions direct the agent to run 'module-local verification commands' and 'wave-level integration checks' which are dynamically sourced from docs/verification-plan.xml.
  • [INDIRECT_PROMPT_INJECTION]: The skill relies on untrusted local files to determine its operational logic and execution parameters. 1. Ingestion points: docs/development-plan.xml, docs/knowledge-graph.xml, and docs/verification-plan.xml (SKILL.md). 2. Boundary markers: No delimiters or ignore-embedded-instructions warnings are defined for the content of these XML files. 3. Capability inventory: The skill has access to shell command execution (via verification steps), file system writes, and git commit operations (SKILL.md). 4. Sanitization: There are no instructions to sanitize or validate the integrity of the commands or architectural constraints read from the documentation before they are acted upon.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 20, 2026, 03:21 PM
Security Audit — agent-trust-hub — grace-multiagent-execute