The Agent Skills Directory

[SAFE]: A thorough audit of the skill instructions and reference documents found no malicious code, obfuscation, or unauthorized network operations. The skill adheres to brand consistency and factual verification.
[INDIRECT_PROMPT_INJECTION]: The skill acts on user-provided text for editing and humanization, creating a surface for indirect prompt injection. \n
Ingestion points: The 'Edit', 'Adapt', and 'Humanize' modes in SKILL.md process external text. \n
Boundary markers: No specific delimiters or "ignore" instructions for user input are defined. \n
Capability inventory: The skill is instructed to run a local linter script (python3 docs/brand/lint.py). \n
Sanitization: No explicit sanitization or filtering of external content is documented. \n
The risk is mitigated by strict semantic-preservation rules and a change-rate guard that stops processing if changes exceed 50%.
[COMMAND_EXECUTION]: The skill requires running a local Python linter (python3 docs/brand/lint.py) as part of its 'Definition of done'. This is a standard project workflow task for verifying documentation against brand contracts and does not involve remote code execution or privilege escalation.

copywriting