skills/compozy/kb/adversarial-review/Gen Agent Trust Hub

adversarial-review

Pass

Audited by Gen Agent Trust Hub on Apr 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to coordinate reviews between external model CLIs and manage temporary files. Evidence includes usage of mktemp, codex exec, and claude -p in SKILL.md.
  • [REMOTE_CODE_EXECUTION]: The skill suggests running tests on code under review via the --profile edit flag in codex exec, which could lead to the execution of malicious code if the input code being reviewed is untrusted.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it interpolates untrusted code diffs into reviewer prompts without adequate boundary protection.
  • Ingestion points: The code or diff to review is inserted into the prompt template in SKILL.md.
  • Boundary markers: Absent; the instructions do not include delimiters or explicit warnings to ignore embedded instructions.
  • Capability inventory: The skill performs shell command execution and file system access as defined in SKILL.md.
  • Sanitization: Absent; the skill does not validate or sanitize input code before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 15, 2026, 12:15 AM