The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts (setup-workspace.sh, check-placement.sh, check-structure.sh) to initialize the environment and verify knowledge base integrity. These scripts use standard system utilities to perform file operations and pattern matching.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documents provided by the user. Evidence chain:\n
Ingestion points: Source documents are ingested during the extraction process described in SKILL.md Step 2.\n
Boundary markers: The skill relies on behavioral instructions (Rule 10 in ba-principles.md) rather than structural delimiters to separate data from instructions.\n
Capability inventory: The skill can execute local bash scripts, run Python code, and write files to the knowledge directory tree.\n
Sanitization: Filename slugging logic in SKILL.md Step 4 explicitly rejects directory traversal components such as .., /, and \.\n- [DYNAMIC_EXECUTION]: The check-frontmatter.sh script executes an embedded Python block via a HEREDOC. The script uses yaml.safe_load() to process user-generated metadata files, ensuring that no arbitrary code is executed during the parsing process.\n- [EXTERNAL_DOWNLOADS]: The skill documentation suggests the user install the pyyaml package via pip3 to enable full metadata validation. The skill performs no automated external downloads and the script handles the absence of the package by skipping the check.

knowledge-curation