The Agent Skills Directory

[SAFE]: The skill implements a set of non-destructive checks for auditing a local wiki. Its operations are scoped to a specific directory structure and it explicitly prohibits automatic modifications of the data it processes.
[DATA_EXPOSURE]: The skill reads local markdown and source files within the wiki/ and raw/ directories. This behavior is standard for its stated purpose of linting a knowledge base. No network exfiltration or access to system-level sensitive files (e.g., credentials, SSH keys) is present.
[COMMAND_EXECUTION]: The skill utilizes standard file system operations like Glob and Grep to search for patterns. It does not invoke arbitrary shell commands or allow user-supplied input to influence shell execution in an unsafe manner.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from wiki pages which are considered untrusted input. While it lacks explicit boundary markers or sanitization for this content, the risk is mitigated by strict internal instructions that mandate a read-only workflow and explicitly forbid the agent from taking any corrective actions automatically.
[INDIRECT_PROMPT_INJECTION]: Ingestion points: wiki/**/*.md, index.md, CLAUDE.md, raw/. Boundary markers: Absent. Capability inventory: Glob, Grep, Read, Write (append-only to log.md). Sanitization: Absent.

wiki-lint