The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for persisting and loading state data via JSON files, which creates an attack surface for indirect prompt injection if the stored data is later interpreted by the agent without sufficient validation.
Ingestion points: Files are read using json.load() in templates/state-manager-template.py and examples/user-state-example.py.
Boundary markers: None; the provided code does not include delimiters or instructions to ignore potential commands within the state values.
Capability inventory: The skill implements file system operations (os.replace, json.dump, fcntl.flock) but does not include network access or arbitrary command execution.
Sanitization: None; the templates focus on persistence mechanics and do not provide specific sanitization for the content of the loaded data.
[DYNAMIC_EXECUTION]: The skill documents a pattern for dynamic library loading to ensure portability across different installation environments.
Evidence: The "Portable Pattern" in SKILL.md dynamically calculates the project root by searching parent directories for a .git folder and modifies sys.path to include plugins/autonomous-dev/lib before performing an import.

state-management-patterns