The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of Harness platform commands via mcp__harness__* tools and the hc CLI fallback for operations like triggering pipelines and managing repositories.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as pull request metadata and pipeline inputs.
Ingestion points: mcp__harness__harness_create (title/body), mcp__harness__harness_execute (inputs), and mcp__harness__harness_diagnose (execution logs).
Boundary markers: None explicitly defined in the prompt instructions to delineate user-supplied data from instructions.
Capability inventory: Full resource lifecycle management (create/update/delete) and triggering pipeline executions.
Sanitization: Instructions rely on platform-level controls; no specific escaping or validation is defined within the skill instructions.
[SAFE]: The skill includes explicit safety rules, such as prohibiting force-pushes to main branches, forbidding the bypass of git hooks, and mandating human confirmation for destructive operations like deleting resources.

harness-automation