The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, such as drafts, proof sources, and publication constraints, which could theoretically contain malicious instructions.
Ingestion points: Defined in the Inputs table in SKILL.md, including 'draft, audience, decision purpose, proof sources, voice rules, and publication constraints'.
Boundary markers: The skill uses 'Decision Rules' and 'Quality Standards' in SKILL.md to distinguish between fact, assumption, and recommendation, and includes a 'Premium Writing Quality Gate' in references/premium-writing-quality-gate.md to verify claims.
Capability inventory: The 'Capability Contract' in SKILL.md restricts the agent to 'read-only for analysis' and 'edit only the requested working copy', explicitly prohibiting publishing, sending, or changing production systems.
Sanitization: The 'Workflow' in SKILL.md includes an 'anti-slop gate' and 'critical-analysis' phase to block release on unsupported claims, acting as a manual review/sanitization step.

premium-commercial-writing