rewrite-natural-prose
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONPERSISTENCE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a bundled Node.js script
scripts/manage-rewrite-policy.mjsto manage its persistence settings. This script uses only standard Node.js built-in modules (fs,path,crypto,url) and does not depend on external packages. - [PERSISTENCE]: The skill is capable of modifying host-specific instruction files (such as
AGENTS.mdfor Codex orCLAUDE.mdfor Claude Code) to enable a continuous rewriting mode. This functionality is transparently documented and includes significant security guardrails: - Token-based Approval: The script generates a unique preview token that binds the specific proposed changes. The agent must present this token and obtain explicit driver approval before any write operation.
- Path Validation: The script strictly validates file paths, requiring absolute paths and refusing to operate on symbolic links to prevent directory traversal or link-following attacks.
- Atomic Writes: It uses temporary files and atomic renaming to ensure file integrity during updates.
- Backup Mechanism: Original instruction files are backed up before modification.
- [DATA_EXPOSURE]: File access is restricted to a set of pre-defined agent configuration filenames within a validated scope root, preventing unauthorized access to sensitive system files.
Audit Metadata