The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a delegation pattern where conversation history is distilled and interpolated into a 'Delegation Packet' used as the initial prompt for a child agent. This architecture presents an indirect prompt injection surface where adversarial content in the conversation could potentially influence the child agent's execution.
Ingestion points: Conversation history is distilled into the 'Established context' section of the packet template in SKILL.md.
Boundary markers: The template uses Markdown headers to organize sections, but lacks explicit delimiters or warnings to the child agent to ignore instructions embedded within the interpolated conversation data.
Capability inventory: The child agent has significant capabilities, including git worktree management, committing code, pushing to remote repositories, and creating pull requests.
Sanitization: No sanitization or filtering of the source conversation is specified before it is passed to the next task.
[COMMAND_EXECUTION]: The skill instructions and the resulting delegation packet describe the execution of shell-level operations, including 'git worktree' commands and running 'repository's relevant tests, checks, and live proof'. These operations are performed within the scope of the agent's primary purpose for code contribution.

delegate