The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text to perform "humanising" edits, which creates a vulnerability to instructions embedded within that data.
Ingestion points: The skill reads external text drafts or files provided by the user using the Read, Grep, and Glob tools.
Boundary markers: The instructions include a "Never invent facts" rule and specific audit questions that serve as semantic checks; however, there are no explicit structural delimiters or markers to isolate the untrusted input from the agent's core logic.
Capability inventory: The skill allows the agent to modify files using the Write and Edit tools and to interact with the user via AskUserQuestion.
Sanitization: No mechanism is specified for sanitizing or escaping potentially malicious instructions within the source text before processing.

humanise