The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/orchestrate.py uses subprocess.run() to manage the workflow by executing other Python scripts within the same package (e.g., graph.py, enforce.py, enrich.py). This orchestration is used for the health and bootstrap commands. The test suite tests/test_autograph.py similarly uses subprocess.run() to validate script outputs against expected behaviors.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading the content of markdown files in a user's vault to perform metadata extraction, tag generation, and link suggestion. The extracted content is sent to LLM providers via the OpenRouter API.
Ingestion points: Vault files are walked and read in scripts/common.py, scripts/enrich.py, scripts/swarm_prepare.py, and scripts/daily.py.
Boundary markers: The scripts/enrich.py script uses file path headers (e.g., ### {path}) to separate multiple note contents within a single prompt, though it lacks explicit instructions to the model to ignore potential instructions embedded within those notes.
Capability inventory: The skill has the ability to modify the local filesystem (writing frontmatter and moving files in scripts/dedup.py), execute local commands (scripts/orchestrate.py), and perform network requests to OpenRouter (scripts/enrich.py).
Sanitization: While frontmatter values are escaped for YAML syntax in scripts/common.py, the body content of notes is treated as raw text when provided to the LLM for analysis.

autograph