The Agent Skills Directory

[SAFE]: The skill is entirely descriptive and provides instructions for a conceptual workflow. It does not include executable scripts, shell commands, or external dependencies.- [PROMPT_INJECTION]: The skill operates by ingesting data from files specified by the user via the --target argument. This presents a surface for indirect prompt injection where instructions hidden within the target files could influence the agent's behavior during the refinement cycle.
Ingestion points: Files or directory paths passed to the --target argument (SKILL.md).
Boundary markers: Not explicitly defined; the skill relies on the agent's internal logic to differentiate between document content and instructions.
Capability inventory: The skill's environment is restricted by the disable-model-invocation: true flag in the YAML frontmatter, limiting the agent's ability to trigger other tools while this skill is active.
Sanitization: No sanitization or validation of the target file content is mentioned in the skill instructions.

Loop