dispatch-weight

Classifies each implementation dispatch as light or heavy to guide per-run model escalation.

2|Updated Jul 24, 2026
One-click install
npx skills add https://github.com/konstantinos-malavazos/claude-code-playbook --skill dispatch-weight-konstantinos-malavazos
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dispatch-weight
Source: https://github.com/konstantinos-malavazos/claude-code-playbook/tree/main/templates/skills/dispatch-weight
Command: npx skills add https://github.com/konstantinos-malavazos/claude-code-playbook --skill dispatch-weight-konstantinos-malavazos

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When an orchestrator hands work to an implementation specialist, it needs a consistent rule for deciding whether that single dispatch should run on a cheaper or stronger model. Without one shared rule, teams either overpay for trivial changes or under-power re-passes that just failed review. ## Core Features & Use Cases - Per-dispatch classification: Emits light or heavy plus a one-line reason naming the criterion that fired, so every dispatch site runs the same auditable rule. - Concrete thresholds: Classifies heavy when a dispatch touches more than 3 files, spans repos, changes a shared contract with multiple callers, works around a placeholder seam, or reopens a design decision on a fix pass. - Three invariants: Enforces raise-on-suspicion asymmetry, a ratchet so re-dispatches never drop below the last tier, and the agent's pinned model: as a floor. - Tier alias mapping: Gives the dispatching agent a table mapping installed model ids to tier aliases (sonnet, opus, haiku, fable), with unmappable ids meaning no override. - Use Case: After a reviewer returns REQUEST CHANGES with a design-level finding, the orchestrator re-classifies the re-pass as heavy and escalates the model for that run only, instead of reusing the original light weight. ## Quick Start Ask the agent to classify the weight of the upcoming implementation dispatch and emit light or heavy with the criterion that fired before dispatching the specialist.

Frequently Asked Questions about dispatch-weight

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I decide which model tier to use for a coding task dispatch?

Classify the single dispatch, not the ticket: heavy if it touches more than 3 files, spans repos, changes a shared contract with multiple callers, or works around a placeholder seam; otherwise light. Emit the weight with a one-line reason naming the criterion that fired.

How should model escalation work on a re-pass after a failed review?

Classify the re-pass on its own merits, then take the higher of that weight and the tier the track last ran at. A finding that the design was wrong, or several findings against one track, makes the re-pass heavy regardless of file count.

Can a per-dispatch weight override an agent's pinned model setting?

A weight can raise a specialist above its pinned model for one run, but never lower it. The agent's frontmatter model is a floor, and the classifier itself never writes or suggests any model value.

What happens when an installed model id does not match a known tier alias?

An unmappable id means no override, never a guess downward. Match on the family word (opus, sonnet, haiku, fable) rather than the exact id, since installers accept arbitrary strings.

Why does a light classification set a tier explicitly instead of no override?

Specialists with no pinned model default to inherit, which is the orchestrator's stronger model. No override would silently hand every light dispatch the expensive tier, so light sets the cheap tier outright.