The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes instructions in SKILL.md and README.md for the agent to execute bundled local scripts (scripts/validate.py and scripts/fact-check-regressions.sh). These tools are used for deterministic validation of the skill's logic, schema compliance, and most importantly, a 'fact-check' mode that prevents the agent from dropping or altering numbers, versions, and technical terms during rewriting.
[SAFE]: The skill is designed with an 'assume-malicious' logic regarding its own model's behavior, implementing multiple safety gates (Gate 1 and Gate 2) and a fidelity verification loop to maintain the integrity of user data. No external network operations, credential harvesting, or obfuscated patterns were detected. All external references in the bibliography are legitimate academic or official sources.

natural-writing