triage

name	triage
description	Decide whether a development task should be done now, handled with `lite-spec`, or sent to `metaplan`.

Triage

Classify an incoming development task into one route — DO_IT_NOW, LITE_SPEC, or METAPLAN — so planning overhead stays proportional to task difficulty. Always prefer the lightest route that still protects execution quality.

When to use

Use this skill when:

• the user asks which planning depth is appropriate
• the task may or may not justify structured planning
• the implementation scope is not yet clear
• the user refers to staged planning, triage, routing, or planning depth selection
• it is unclear whether immediate execution would be efficient or reckless

Do not use this skill when the user has already chosen a specific planning mode and that choice is clearly appropriate.

Decision model

Assess the task on five axes. Each row names the axis's risk end and what it means; the safe end is the converse (concrete / local / easy to undo / unlikely to pause / planning overhead exceeds likely rework cost).

Axis (risk end)	Risk end means
Ambiguity (high)	requirements, scope, constraints, or completion criteria read in more than one materially different way
Blast radius (high)	crosses architecture, interfaces, invariants, or migration paths
Reversibility (low)	a wrong implementation is hard or costly to undo
Interruption risk (high)	a coding agent would likely stall, guess, or re-research
Review value (high)	upfront planning would materially cut rework, drift, or hallucinated decisions

Routing procedure

Evaluate in order; the first match wins.

1. METAPLAN — Ambiguity high, Blast radius high, Reversibility low, or Interruption risk high (any one), or execution depends on a hardened spec/plan/task breakdown that does not yet exist. Review value at its risk end alone, with the other four axes safe, does not trigger METAPLAN.
2. DO_IT_NOW — otherwise, if most axes are at their safe end.
3. LITE_SPEC — otherwise (default): bounded but not trivial; a short brief would reduce drift.

Tie resolution

A bounded task can also trip a METAPLAN trigger; the ordering resolves this: METAPLAN wins, because material ambiguity, invariant impact, costly-to-undo mistakes, and likely stalls outrank scope-boundedness. Below that bar, LITE_SPEC remains the default for mixed signals; route DO_IT_NOW instead only on a genuine borderline tie (unclear whether most axes sit at their safe end) where a brief would merely restate an already concrete task — if a brief would capture any constraint, risk, or edge case worth writing down, LITE_SPEC stands.

Insufficient information

Treat an unassessable axis as ambiguity. Ask one targeted question naming what is missing only if the answer could change the route; otherwise assume the heavier plausible reading and proceed.

Output format

Return:

1. the selected route
2. a short justification
3. the main factors that drove the decision (axis levels from the Decision model, not restated criteria)
4. the immediate next action

Use this structure:

Route: <DO_IT_NOW | LITE_SPEC | METAPLAN>

Why:
- ...
- ...

Key factors:
- Ambiguity: low|medium|high
- Blast radius: low|medium|high
- Reversibility: high|medium|low
- Review value: low|medium|high

Next action:
- ...

Boundary cases

• DO_IT_NOW: a wording fix or narrow bug fix with obvious acceptance criteria.
• LITE_SPEC: a feature spanning several files in one subsystem, bounded but not trivial.
• METAPLAN: a migration touching a public interface with several unresolved design choices.
• LITE_SPEC/METAPLAN boundary: a multi-file refactor with clear scope, no material ambiguity, low interruption risk, moderately costly worst case → LITE_SPEC (no trigger fires); the same refactor with one API decision left open and a hard-to-undo migration → METAPLAN (Interruption risk high and Reversibility low both trigger).
• Multi-condition: Blast radius and Review value at risk ends, Ambiguity low, Reversibility high → METAPLAN; Blast radius alone triggers it (step 1 matches first).
• Insufficient information: "improve the auth flow" with no target behavior stated → the missing target could shift the route between LITE_SPEC and METAPLAN; ask one question naming it before routing.
• Lightest-route bias: a mechanical, reversible rename touching six files, no interface change → DO_IT_NOW, not LITE_SPEC; file count alone does not put Blast radius at its risk end, and a brief would only restate the task.

Important constraints

• Prefer the lightest adequate route; never escalate merely because the task is technical.
• Do not under-plan when ambiguity would predictably cause clarification loops or rework.
• Optimise for total execution efficiency, not procedural formality.