The Agent Skills Directory

[COMMAND_EXECUTION]: The lib/entity_resolver.py file uses subprocess.run to invoke the ripgrep (rg) utility for searching entity references. Evidence: The skill executes subprocess.run with a list of arguments and shell=False (default). Context: The search query is rigorously sanitized using a regex allow-list (alphanumeric and limited symbols) and is passed to ripgrep with the -F flag (fixed string) to prevent regex injection attacks.
[EXTERNAL_DOWNLOADS]: The lib/ontology_registry.py file includes functionality to fetch ontology definitions from remote URLs. Evidence: The load_from_url method uses urllib.request.urlopen to download YAML content. Context: The implementation restricts allowed URL schemes to http and https and processes the downloaded content using yaml.safe_load, which prevents the execution of arbitrary Python objects during parsing.
[PROMPT_INJECTION]: The skill's entity indexing mechanism processes memory files (*.memory.md), which constitutes an indirect prompt injection surface. Ingestion points: lib/entity_resolver.py reads the content of memory files during indexing. Boundary markers: Boundary markers (delimiters) are absent in the processing logic for memory content. Capability inventory: The skill's capabilities include executing shell commands via ripgrep and performing standard file operations. Sanitization: Frontmatter content is parsed using yaml.safe_load to ensure that data structures are handled safely.

ontology