The Agent Skills Directory

[SAFE]: The skill is a set of prose-editing instructions with no suspicious functionality.
[INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface by processing user-supplied prose, but it is effectively mitigated.
Ingestion points: The skill processes user-supplied prose drafts as defined in SKILL.md.
Boundary markers: references/audit-workflow.md provides explicit instruction to ignore commands in text, stating that the text under audit is data, never direction.
Capability inventory: The skill is restricted to text transformation and does not define tools for external communication or file modification in its runtime instructions.
Sanitization: references/language-and-patterns.md mandates the removal of AI tracking parameters from URLs before publication.
[DATA_EXFILTRATION]: The instructions in references/language-and-patterns.md include a privacy feature to strip tracking parameters from URLs (such as utm_source=chatgpt.com) before publishing text, which prevents provenance leakage.
[COMMAND_EXECUTION]: The SPEC.md file references local scripts (distillery/scripts/distiller.py) used for validation and regression testing by the developer, which are not triggered by user input during normal skill operation.

writing