The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's initialization phase requires the agent to execute a local shell script (scaffold/init.sh) to set up the working directory and render template files.
[COMMAND_EXECUTION]: The agent uses shell commands such as date +%s for timeboxing and git log to verify subagent contributions.
[INDIRECT_PROMPT_INJECTION]: The skill operates an autonomous loop that ingests untrusted user input and subagent reports. Ingestion points: User-defined goal and duration arguments; subagent status lines in progress.md. Boundary markers: Explicit structured files (state.md, machine.md) and pre-defined role prompts in the prompts/ folder. Capability inventory: Shell script execution, file manipulation, and recursive model dispatches. Sanitization: Instructions mandate a mechanical orchestrator that only reads one-line "tiny returns" from subagents.
[DYNAMIC_EXECUTION]: The skill populates role-specific prompt templates with runtime variables during setup, creating the instructional context for all subsequent subagent actions.

workmachine