The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes durable goal state stored in JSON format, which includes user-provided objectives, constraints, and evidence. * Ingestion points: The agent is instructed to read and resume goal state from workspace files like templates/goal.json and examples/migration-goal.json. * Boundary markers: Uses a strict JSON schema and explicit instructions to separate goal tracking from ordinary task execution. * Capability inventory: The skill has the capability to write to the local workspace and execute validation scripts. * Sanitization: The scripts/validate_goal.py script performs deep validation, rejecting non-finite numbers and duplicate keys, and enforcing type safety.
[DYNAMIC_EXECUTION]: The unit test script scripts/test_validate_goal.py uses dynamic loading to import validation logic for testing. * Evidence: Employs importlib.util.spec_from_file_location to load the scripts/validate_goal.py module from a relative file path.
[COMMAND_EXECUTION]: The skill executes its own bundled Python script to verify the integrity of goal contracts. * Evidence: SKILL.md directs the agent to use python3 scripts/validate_goal.py <goal.json> to validate state before pursuit.

agentic-goals