The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands (grep, echo, kill) to manage project configuration and session locking. Specifically, it ensures the .harnessed/ directory is added to .gitignore and uses kill -0 to verify if a process associated with a lock file is still active. These are standard, local utility operations.
[SAFE]: The skill processes user tasks to generate QA artifacts, creating a surface for potential indirect instructions.
Ingestion points: User-provided task descriptions are used to generate acceptance criteria in .harnessed/contract.md (SKILL.md).
Boundary markers: The skill employs <HARD-GATE> markers to enforce adherence to the QA pipeline (SKILL.md).
Capability inventory: The skill can execute shell commands for file management and process checking, and it can write and rename files in the .harnessed/ directory (SKILL.md, reference.md).
Sanitization: No specific sanitization or escaping of user-provided task content is mentioned before it is processed into artifacts.
[SAFE]: The skill includes a 'High-Risk Task Detection' mechanism that identifies sensitive operations, such as handling secrets or authentication, and mandates a more rigorous QA process, which serves as a defensive measure.

using-harnessed