verification-before-completion

Pass

Audited by Gen Agent Trust Hub on Aug 10, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill provides instructions to clone the vendor's GitHub repository and execute internal shell scripts (validate-skills.sh, check-marketplace.sh) to perform skill validation. This pattern involves the execution of remote code managed by the skill author.
  • [INDIRECT_PROMPT_INJECTION]: The skill mandates that the agent ingest and evaluate output from various external tools, such as linters, test suites, and build commands, to verify task completion. This creates a surface where untrusted data could influence agent behavior.
  • Ingestion points: Terminal output from bash verification commands and reports from the OMO review-work skill.
  • Boundary markers: Absent; the skill directs the agent to read the 'full output' of commands without specific delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill is permitted to use Bash, Read, Glob, and Grep, providing significant file system and execution capabilities.
  • Sanitization: None; the skill relies on the agent to interpret raw tool output to verify claims.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 10, 2026, 08:13 AM
Security Audit — agent-trust-hub — verification-before-completion