skills.rest
reddb-io
reddb-ioOFFICIAL
Explore all skills by this author →
GitHub

model-tier-policy

Routes AFK tasks to model tiers for validation, simple code, complex code, and design.

20|2|Updated May 16, 2026
Search Tags:#claude code#codex#cost optimization#model routing#tier classification#afk execution#llm policy
One-click install
npx skills add https://github.com/reddb-io/red-skills --skill model-tier-policy-reddb-io
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: model-tier-policy
Source: https://github.com/reddb-io/red-skills/tree/main/plugins/dev/skills/engineering/model-tier-policy
Command: npx skills add https://github.com/reddb-io/red-skills --skill model-tier-policy-reddb-io

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Choosing the right model tier for each engineering task is hard to keep consistent across Claude Code, Codex, and AFK execution, and copying tier tables into executor prompts causes the classification criteria to drift. This Skill centralizes the RedSkills dev model tier policy so every task runs on the cheapest capable tier and escalates on evidence. ## Core Features & Use Cases - Tier classification policy: Defines the validate, simple, complex, and think tiers with default Claude, Codex, and MiniMax model/effort values and concrete classification criteria. - Deterministic-first validation: Directs structural checks (schema, lint, tests, type checks) to run before spending any model tokens on fuzzy validation. - Escalation and runtime overrides: Documents evidence-based escalation between tiers and the --model/RED_AFK_MODEL override precedence for ad-hoc and CI runs. - Use Case: When triaging an issue for autonomous execution, consult this policy to decide whether it is a simple single-scope change or a complex cross-module change, then let AFK route the tier to the right model. ## Quick Start Ask the agent to classify an issue using the model tier policy and explain which tier should execute it and why.

Frequently Asked Questions about model-tier-policy

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

FAQPage Schema
How do I choose a model tier for an AFK task?▼

Classify the task against the tier table: validate for fuzzy semantic checks, simple for well-specified single-scope code, complex for cross-module or high-blast-radius changes, and think for design or planning. Start with the cheapest capable tier and escalate on evidence.

What is the difference between simple and complex code tiers?▼

Simple applies when the request is well specified, single-scope, avoids architecture or contract changes, and has a narrow deterministic validation path. Complex applies when work crosses module or runner boundaries, changes public contracts or schemas, or carries security, data, or concurrency risk.

Can I override the model tier at runtime without editing config?▼

Yes. The --model flag and RED_AFK_MODEL environment variable override .red/config.yaml and defaults, with the same precedence for --effort and RED_AFK_EFFORT. A non-empty override flattens every tier onto the one model slug.

Does the model tier policy work with Codex interactive sessions?▼

Codex interactive runs a single session model and cannot pin per-call models yet. The tier still applies to Codex through AFK tier-routing, where the codex runner adapter resolves the per-issue tier into the spawn's --model and --effort flags.

When should deterministic checks run before the validate tier?▼

Deterministic tools should run first for any structural question: schema validation, JSON/YAML parsing, shell syntax, type checks, lint, tests, and exact file checks. Reserve the validate tier for fuzzy semantic judgment that deterministic checks cannot answer.

Related Skills

Explore More Software Engineering Skills→
andrewthetechieandrewthetechie
2

issue-triage

Classifies incoming issues and produces a structured Markdown triage artifact.

Community
Advanced
cody-hutsoncody-hutson

pmo-tier-1-support

Classify and resolve known support issues while escalating novel problems.

Community
Intermediate
dariovr1dariovr1
1

configuring-sixth-sense

Configures model preferences and effort levels for Sixth Sense subagent dispatch.

Community
Basic
0xmariowu0xmariowu
40

autosearch:model-routing

Route leaf skills to optimal model tiers using frontmatter model_tier annotations.

Community
Basic