The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run() in scripts/orchestrate.py and tests/test_autograph.py to coordinate the execution of its various internal Python utility scripts. While the commands are constructed as lists and do not use a shell, this pattern involves dynamic invocation of local executables.
[EXTERNAL_DOWNLOADS]: In scripts/enrich.py, the skill makes network requests to openrouter.ai to perform LLM-based tagging and link enrichment. This involves sending vault content to an external service. openrouter.ai is a well-known technology service for LLM access.
[DYNAMIC_EXECUTION]: The script scripts/common.py uses yaml.load() with a custom FrontmatterLoader that inherits from yaml.SafeLoader. This is a safe implementation that prevents arbitrary code execution through YAML tags while allowing the parsing of complex vault metadata.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown data from an Obsidian vault to extract entities, decisions, and relationships.
Ingestion points: scripts/daily.py, scripts/enforce.py, and scripts/graph.py read the content of markdown files within the user's vault.
Boundary markers: The skill does not consistently apply clear delimiters or 'ignore' instructions when presenting extracted vault content to the agent or LLM for review.
Capability inventory: The skill possesses capabilities to write to the file system (Path.write_text) and execute internal scripts via subprocess.run.
Sanitization: Metadata is sanitized via a SafeLoader, but the natural language body of the notes is processed primarily through regular expressions and direct inclusion in LLM prompts.

autograph