The Agent Skills Directory

[SAFE]: The skill implements a robust verification sequence (Inspect, Prepare, Verify, Apply, Verify Result, Complete) for all workspace modifications, requiring the agent to reread and verify file bytes before finalizing changes. This significantly reduces the risk of accidental or malicious data corruption.
[SAFE]: The provided Python script (scripts/update_preference.py) manages local configuration files using safe practices, including checks to prevent symlink-based attacks and the use of restrictive file permissions (0600).
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through workspace administration tasks, but the risk is managed via the primary skill instructions.
Ingestion points: The skill reads STUDIO.md, PROJECT.md, and TASKS.csv to synchronize status and manage project registration.
Boundary markers: Not explicitly defined in the provided instruction set.
Capability inventory: The skill is allowed to use Write, Edit, and Bash tools as defined in the YAML frontmatter.
Sanitization: The skill mandates a 'native mutation sequence' where the agent must establish durable saving and inspect actual permissions and ownership before the final write operation.

studio