The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to verify documentation claims using the "Shell in non-mutating or dry-run mode" and by executing "a safe inspection command when text is insufficient." Specifically, it directs the agent to "Execute examples only in a safe local or disposable context" to verify parsing and expected output. This capability can be exploited if the documentation contains malicious commands disguised as benign examples.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository files and external documentation sources, which creates a vulnerability to embedded instructions.
Ingestion points: Documentation entrypoints, public docs, maintainer docs, operational runbooks, and code comments are identified as scope in the checklist.
Boundary markers: The instructions mandate defining the "evidence question" before retrieval and using tools for "mechanical processing" to "summarize bounded results," keeping "full output outside context."
Capability inventory: The skill allows for local shell command execution, file system reads, and web-based research of official specifications.
Sanitization: The skill relies on "mechanical processing" and summarization as a defense, though it does not specify explicit character escaping or input validation before interpolating content into the agent's context or shell commands.

ln-53-documentation-auditor