Goal Loop

Goal Loop

Use this skill when the user has a North Star but does not yet know the best path.

When to use

• The user wants to iterate toward a goal through repeated bets rather than follow a fixed implementation plan.
• The work is open-ended optimization, investigation, or metric improvement.
• Restart safety and compounding learning matter more than a one-shot plan.

When not to use

• The work already has a fixed-scope plan. Use arch-step, arch-mini-plan, or lilarch.
• The task is a concrete bug investigation or fix. Use bugs-flow.
• The user needs a math-first Commander’s Intent investigation specifically. Use north-star-investigation.

Non-negotiables

• The controller doc and append-only worklog are first-class. Read the worklog before every iteration.
• New or draft goal-loop docs require North Star confirmation before the loop becomes active.
• One iteration means one bet with a pre-committed decision rule.
• No reruns of the same bet without changing a lever or evidence surface.
• Setup mode is docs-only. Code changes only happen during iteration mode.
• Keep the loop honest about the best current belief and biggest uncertainty.

First move

1. Read references/controller-contract.md.
2. Read references/shared-doctrine.md.
3. Resolve the mode:
   • new or repair controller
   • readiness / flow-status
   • iterate
4. Resolve DOC_PATH and WORKLOG_PATH.
5. Read the most recent worklog entries before choosing a new bet.
6. Read the matching mode reference and references/quality-bar.md.

Workflow

1) Setup or repair

• Create or repair the controller doc and the append-only worklog.
• Draft the North Star, scoreboard, lever inventory, and anti-sidetrack contract.
• Pause for North Star confirmation when the doc is new or still draft.

2) Flow-status

• If the user asks what is next, inspect the controller and worklog for readiness.
• Recommend the single best next move rather than reciting the whole doctrine.

3) Iterate

• Re-read the controller contract and the latest worklog entries.
• Choose one bet and pre-commit the decision rule.
• Do the minimum work needed for that bet.
• Run the smallest credible signal.
• Append exactly one worklog entry with evidence, learnings, and the next bet.

Output expectations

• Update DOC_PATH and WORKLOG_PATH when the loop state changes.
• Keep the console output short:
  • North Star reminder
  • punchline
  • current best belief
  • biggest uncertainty
  • next bet
  • pointers to the controller/worklog

Reference map

• references/controller-contract.md - required blocks, worklog rules, status rules, and loop contract
• references/shared-doctrine.md - loop discipline, bet quality, and anti-sidetrack rules
• references/setup.md - create or repair the controller and make the loop ready
• references/iterate.md - choose one bet, pre-commit the rule, and append one honest worklog entry
• references/flow-status.md - read-only readiness and next-bet recommendation rules
• references/context-digest.md - write or refresh the restart-safe digest only when requested
• references/quality-bar.md - strong vs weak controller state, bets, and worklog entries