Can I state "done" in one sentence?
The result must be checkable. If the goal is still vague, implementation is premature.
A task-entry map for deciding what to do next. It assumes many tasks start in unfamiliar fields, so unknown-domain work is routed into controlled discovery and reversible automation.
The result must be checkable. If the goal is still vague, implementation is premature.
The task is not ready to delegate as implementation.
The outcome is checkable enough to route.
Classify the current step, not the whole project forever.
The real question or goal is unclear.
I need facts, cause, options, or examples.
I need code, a report, a workflow, or another artifact.
I need defects found or parts combined.
This determines how much authority the agent is allowed to have.
I understand the domain well enough to specify and judge the result.
I can learn enough with focused investigation and review.
I cannot reliably judge final correctness yet.
Unknown-domain mode is for moving carefully, not stopping.
The work maps, compares, formats, retrieves, checks consistency, drafts, or flags anomalies.
The work approves, submits, deletes, publishes, changes shared state, or decides correctness.
Containment comes before implementation, not after.
It affects production, shared infrastructure, money, deletes, sends, publishing, or approvals.
The action is reversible, isolated, or only produces a draft/artifact.
Default to the simplest structure that fits. Split only when the reason is concrete.
One domain, one output, one sitting of work.
Later work needs earlier answers or artifacts.
Several investigations do not share state or edit the same artifact.
Multiple agents may write at once.
This is the most expensive mistake, so the bar is high.
I cannot name disjoint files, isolate runtime state, test each part, or plan integration.
Files do not overlap, runtime is isolated, tests are independent, and integration is planned.
Choose the job shape first. Tool availability can change.
Framing, quick analysis, critique, or writing.
Multi-source evidence needs citations and a reviewable plan.
Documents, data, implementation, debugging, refactors, or tests.
Websites, private tools, or authorized workspace data are required.
Do not accept pasted confidence. Require proof I can inspect or rerun.
Evidence checked, source locations, confidence, unknowns.
Diff, exact files, assumptions, exact verification steps.
Exact defect location, severity, proven issue versus risk.
Conflicts, resolutions, and one end-to-end check.
The result is not done until the proof gate survives verification.
Evidence is inspectable and the result matches the one-sentence goal.
The agent missed the gate but the structure still looks correct.
The loop is stuck, underspecified, or in the wrong mode.
These are routing defaults, not rankings. Plan and workspace availability vary. See the interactive version for dated official source notes and mode-specific recommendations.