The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, specifically GitHub Actions logs and diagnostic classifications from the StarSling control plane.
Ingestion points: Data enters the context via sling logs and the prompt field in the sling why command output.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing tool-provided diagnostic strings.
Capability inventory: The agent can perform significant repository actions via gh including re-running jobs, cancelling runs, triggering workflows, and managing secrets or variables.
Sanitization: There is no mention of sanitizing or validating the prompt field returned by the remote API before the agent is instructed to act on it.
[COMMAND_EXECUTION]: The skill coordinates the execution of the sling and gh command-line tools.
Environment Verification: The skill uses sling doctor to verify authentication, connectivity, and tool health before performing operations.
Command Safety: It uses a predefined command surface and routing table to prevent the agent from inventing or executing unsupported subcommands.

sling