The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from URLs and external documents during the ingest operation, which could contain instructions intended to influence agent behavior during summarization or cross-referencing.
Ingestion points: Uses the web_extract tool to fetch external content from URLs and save it to the raw/ directory (described in SKILL.md).
Boundary markers: The instructions do not specify any delimiters or safety warnings to wrap external content when the agent processes it for wiki updates.
Capability inventory: The skill utilizes read_file, write_file, search_files, and patch tools for file management, and executes a local Python script for auditing.
Sanitization: There is no evidence of sanitization or filtering of the ingested content before the agent analyzes and incorporates it into wiki pages.
[COMMAND_EXECUTION]: The skill executes a local Python script to perform maintenance tasks such as finding broken links and orphaned pages.
Evidence: SKILL.md contains instructions to execute python3 ~/.hermes/skills/md-wiki/scripts/lint_wiki.py with path arguments.
Context: The script scripts/lint_wiki.py uses standard libraries (os, re, argparse) and only performs file system operations within the specified wiki directory.

md-wiki