The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a durable state mechanism that reads from a local file (.use-goal/current.md). This file acts as an ingestion point for data that influences agent behavior during the autonomous execution loop.
Ingestion points: The agent reads the Objective, Definition of done, and Next action fields from .use-goal/current.md whenever the goal is resumed or reconciled.
Boundary markers: The skill uses a structured markdown format and includes logic to detect data collisions, but it lacks explicit instructions to treat the natural language within the state file as untrusted or to ignore embedded instructions.
Capability inventory: The skill allows the agent to perform autonomous work, including file inspection, code editing, diagnostic fixes, and test execution.
Sanitization: The skill performs workspace reconciliation to ensure code state matches the checkpoint, but it does not sanitize or validate the natural language instructions stored in the state file before execution.
[PERSISTENCE]: The skill establishes persistence by creating and maintaining a hidden directory .use-goal and state file current.md within the user's workspace to track goals across sessions. While this is the primary purpose of the skill, it is a persistent modification to the environment.

use-goal