The Agent Skills Directory

[COMMAND_EXECUTION]: The skill incorporates a local Python script (scripts/audit-rewrite.py) which is designed to be executed by the agent to verify the integrity of rewritten content. The script uses only Python standard libraries to perform deterministic regex-based analysis, ensuring that protected elements (like URLs, code snippets, and numerical values) are not altered during the rewriting process. The invocation method uses standard CLI patterns and avoids dangerous practices like shell injection.
[INDIRECT_PROMPT_INJECTION]: As the skill involves rewriting user-supplied drafts, it has an inherent attack surface for indirect prompt injection. However, the skill implements specific mitigations, including a 'Hard Contract' for the AI and an audit script that explicitly detects and flags 'prompt residue' such as 'ignore previous instructions' or model-specific system markers. This proactive detection reduces the risk of malicious instructions in the source data affecting the agent's behavior.
[SAFE]: The skill demonstrates a strong security posture by explicitly refusing to engage in deceptive practices like detector evasion or the fabrication of human personas. It relies on a 'protected-content ledger' to maintain factual and structural fidelity. All external resources and bibliography links point to reputable and well-known services (e.g., W3C, Microsoft, Google, Wikipedia), and the author identity is consistent with the provided installation paths.

convert-to-natural-writing