convez.ai
Skill
milkoor

prompt-routing-skill

milkoor/prompt-routing-skill

Ver no GitHub
InicialPluginScore · de 0 a 100

Prompt posture router for coding tasks; pairs with causetrace

Instalação

$npx convez-ai milkoor/prompt-routing-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
0

README

A tradução para o português deste README ainda está na fila.

Prompt Routing Skill

A lightweight skill for choosing the right prompt posture before a coding task starts.

This skill was derived from causetrace research on prompt morphology in real development workflows. causetrace measures the runtime effects; this skill applies the routing decision up front.

Labels

  • prompt-routing
  • routed-prompt
  • causetrace-prompt-posture

Lane Boundary

This skill creates a workflow intervention lane.

  • direct-prompt native trace stays the native baseline
  • routed-prompt trace is not a natural prompt and should not be merged into direct-native conclusions
  • expanded prompt study trace belongs to the controlled / intervention lane
  • external trajectory stays separate

What it does

  • minimal_prompt for open-ended or exploratory work
  • human_structured_prompt for medium-complexity tasks
  • expanded_constrained_prompt for safety-sensitive or boundary-sensitive tasks

Causetrace Integration

This skill emits explicit causetrace_tags so causetrace can classify routed-prompt traces into the routed_prompt_intervention lane without heuristic inference.

Each routing decision includes a structured metadata block:

causetrace_tags:
  - prompt-routing
  - routed-prompt
  - causetrace-prompt-posture
intervention_lane: routed_prompt_intervention
intervention_evidence_level: strong
intervention_evidence_source: prompt-routing-skill
prompt_posture: <selected posture>
routing_decision_id: <stable id>
selected_prompt_style: <style>
routing_reason: <brief reason>

This keeps the routed_prompt_intervention lane auditable and prevents heuristic lane contamination.

Supported agents

  • Claude Code
  • Codex
  • OpenCode

Related project

This skill is designed to pair with causetrace:

  • causetrace records and analyzes the resulting runtime morphology
  • prompt-routing selects the prompt posture before execution

The two repositories are intended to reinforce each other:

  • prompt-routing improves the quality and consistency of task framing
  • causetrace measures how that framing changes execution traces

Install

Run the bundled installer:

./install.sh

Or copy the skill into your agent's skills directory.

Files

  • SKILL.md - skill instructions
  • agents/openai.yaml - UI metadata
  • install.sh - installs the skill into local agent skill directories

License

MIT

Ver o original no GitHub

Bom para

Relacionados