The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted user-supplied English text, creating a surface for indirect prompt injection attacks where malicious instructions might be hidden within the content to be "humanized."
Ingestion points: The primary input text processed by the skill as described in SKILL.md (Received text).
Boundary markers: The skill contains an explicit "Security boundary" section in SKILL.md that defines all target text and embedded links as untrusted data rather than instructions.
Capability inventory: The skill utilizes file system access, network access for citation verification (e.g., doi.org), and the ability to execute regex-based analysis through shell or Python.
Sanitization: The instructions specifically mandate that the agent should ignore embedded commands (e.g., "ignore the above") and treat them as "tells" of AI generation instead of actionable instructions.
[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to generate and execute "throwaway" Python scripts to facilitate complex linguistic analysis and regular expression matching.
Evidence: SKILL.md suggests using Python scripts for sweeps where standard grep utility might be insufficient, specifically noting that these should only use the Python standard library.
[COMMAND_EXECUTION]: The skill utilizes command-line utilities such as grep and the Python interpreter to scan documents for patterns defined in its reference catalogs.
Evidence: Technical instructions in SKILL.md detail how to check for and use python3 or py -3 depending on the host operating system.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to connect to external services to verify the validity of academic citations and links found in target text.
Evidence: references/source-fabrication.md directs the agent to verify DOIs via doi.org and ISBNs via worldcat.org, which are established, well-known services.

awesome-humanize-en