write-hook

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of hook scripts that ingest data from stdin and inject context back into the agent. This represents an indirect prompt injection surface where maliciously crafted input from the environment or previous turns could influence the agent's behavior. Ingestion points: Hook scripts read JSON data from stdin (e.g., INPUT=$(cat) in SKILL.md). Boundary markers: The skill provides a stop_hook_active guard to prevent infinite loops but does not explicitly specify delimiters for the injected additionalContext to prevent instruction injection. Capability inventory: The skill involves writing and executing shell scripts (e.g., bash, jq, git) which have full access to the project environment and can perform file operations (touch). Sanitization: The skill emphasizes the use of stop_hook_active to prevent recursive turn restarts, which is a key reliability and safety mechanism for this platform.
  • [COMMAND_EXECUTION]: The skill provides templates and instructions for creating shell scripts to be executed as hooks by the agent harness. This includes using tools like grep, sed, jq, and git to process data and interact with the filesystem.
  • [PRIVILEGE_ESCALATION]: The skill instructions and associated evaluation transcripts include using chmod +x to grant execution permissions to dynamically created hook scripts. While this is necessary for the intended functionality, it involves the modification of file permissions to allow script execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 07:27 AM