The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references technical documentation and academic research papers from reputable domains such as arxiv.org and learn.chatgpt.com. These links are used for educational context regarding scheduling theory and subagent configuration.
[COMMAND_EXECUTION]: The skill manages the execution of subagents and persists state using local file operations. It specifically creates and updates TASKS.md and JOURNAL.md files within a workspace-specific .divisible-work/ directory to track progress and decisions.
[INDIRECT_PROMPT_INJECTION]: As an orchestration tool, the skill ingests data produced by autonomous subagents, which represents a potential attack surface.
Ingestion points: Result contracts, artifacts, and verification evidence from subagents are processed by the coordinator as described in SKILL.md.
Boundary markers: Assignments include mandatory role boundary instructions that restrict worker behavior and prevent unauthorized recursive calls or ledger edits.
Capability inventory: The skill utilizes subagent spawning, model/effort routing parameters, and local file read/write operations.
Sanitization: The coordinator is instructed to verify all evidence against pre-defined acceptance criteria and delegate verification to separate agents for critical outputs.

divisible-work