looplisted

# CGEL loop Work inside a sealed task proceeds in small, evidence-checked iterations. The control layer counts everything; your job is to make each iteration a real experiment — and to spend roundtrips on work, not ceremony. When more than one task is open, add `--task <id>` to every command below and only ever decide the task this session owns. ## The cycle 1. **INVESTIGATE** — read enough to form a hypothesis about what to change and why it will meet a criterion. For broad reading in big codebases, use the read-only `cgel:explorer` subagent instead of flooding context. 2. **PLAN ITERATION** — open it explicitly: `cgel iterate open --hypothesis "H-1: ..." --change "smallest change that tests it" --expect unit-tests` (`--change` = `--intended-change`, `--expect` = `--expected-checks`.) Always declare `--expect`: an iteration that claims nothing cannot ADVANCE. The first open moves the task SEALED → ACTIVE. You cannot open a second iteration while one is undecided. 3. **CHANGE** — edit inside `scope.allowed` only. Mechanical bulk edits can go to the `cgel:worker` subagent with an exact spec; decisions stay here. Build to the production bar, especially in a messy codebase: reuse existing helpers instead of duplicating them, fix the root cause instead of papering over it (CGEL-ROOT-1 BLOCKS), catch the null-check and edge case in your own diff (CGEL-CORRECT-1 BLOCKS), and update every caller/test/doc the change touches (CGEL-IMPACT-1 BLO