Deepen a Task

Improve the approach, then do the work. Deeper, not bigger, is the governing principle. Respect requests limited to planning, exploration, or prompt editing.

Anchor

Inspect the relevant context. Identify the intended outcome, fixed constraints, and observable success condition. Distinguish requirements from assumptions about how to achieve them. If no task is identifiable, ask for it.

Apply the useful lenses

Consider all three; use only what materially improves the task.

• Altitude: What should this task enable? Challenge the proposed means, not the user's goal. Prefer a better route within scope; seek approval for material changes to scope, cost, risk, or commitments. Preserve explicit deliverables and creative intent.
• Depth: What would a leading practitioner do differently? Start from primary evidence, challenge consequential assumptions, and use concrete, reproducible methods—not expert vocabulary. Where useful, test a specific stretch target rather than assuming the current limit is necessary.
• Range: Describe the task with verbs rather than artifact names. Borrow a technique from another field that addresses a specific weakness. Apply the method; don't role-play experts or automatically assemble agents.

Prefer the simplest approach that improves the result. More tools, analysis, and deliverables are not inherently more ambitious.

Execute and verify

Briefly explain the chosen improvement, then execute within the authorized scope. Start with a bounded step that produces value or tests the riskiest assumption; continue through verification rather than stopping at a plan.

Check the result against evidence outside your own judgment: reproducible tests, held-out data, measurements, authoritative records, or independent review. State what would falsify the important claims.

Report the result, checks actually completed, and remaining uncertainty. Never imply unavailable evidence was inspected or proposed validation ran. Stop at the agreed outcome or a genuine blocker—not adjacent work.