The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/new_session.py executes a git command using subprocess.run with a list of arguments. This method is secure against command injection as it avoids shell interpretation of the parameters.- [DATA_EXFILTRATION]: Analysis of the Python scripts confirms that data access is limited to the local project harness directory and standard Git metadata. There are no functions performing network requests or external data transfers.- [PROMPT_INJECTION]: The skill processes user-provided inputs like session scope and outcomes which are interpolated into markdown files. This behavior represents a standard documentation use case.
Ingestion points: Script arguments --scope, --agent, and --outcome in scripts/new_session.py and scripts/close_session.py.
Boundary markers: Data is structured using markdown headers and YAML frontmatter delimiters.
Capability inventory: The skill can read/write local files and execute Git commands.
Sanitization: Filenames and identifiers are processed through a slugify function, while content fields are written as provided.

harness-session