The Agent Skills Directory

[SAFE]: The skill's primary purpose is academic text humanization, specifically removing AI-like defensive prose and boilerplate from research manuscripts and experiment plans to improve clarity and professionalism.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection as it processes untrusted research artifacts provided by users.
Ingestion points: Manuscript drafts, result tables, method descriptions, and publication files (SKILL.md, workflow step 1).
Boundary markers: The skill uses a defined "Warning Contract" and "Output Contract" to ensure that meta-commentary and safety concerns are reported out-of-band to the user rather than being silently injected into documents (SKILL.md, sections 'Warning Contract' and 'Output Contract').
Capability inventory: The skill performs text manipulation and hands off execution to other domain-specific skills such as ccf-paper-writer and ccf-experiment-designer (SKILL.md, invocation controls).
Sanitization: Content classification is handled through specific local policy references (references/humanization-policy.md) which guide the agent on what to remove, keep, or warn about.
[EXTERNAL_DOWNLOADS]: The skill references local configuration and shared policies located in a relative sibling directory (../ccf-common/). No external network requests, URL fetching, or remote script downloads are present.

ccf-humanization