Research Tick

Execute one iteration of the research loop. This skill is called repeatedly — either by the daemon, by /loop, or manually.

Process

1. Read the active research brief from .loop/briefs/ (the most recent research-* brief with status "running")
2. Read the research module config from .loop/modules/research/config.json
3. Dispatch a research worker using the Agent tool:
   • Use subagent_type: "general-purpose"
   • Pass the research worker agent prompt from the module
   • Include paths to: brief, findings.md, sources.json, coverage.json, search-log.jsonl
   • The worker performs one action and updates state files
4. Check if evaluation is due (every eval_interval iterations):
   • If yes, dispatch the evaluator agent
   • Evaluator updates coverage.json and writes to eval-log.jsonl
   • If evaluator says STOP, mark the brief as "complete"
5. Check iteration count against max_iterations