agentleFS
ForgeyClapForgeyClap/Forge-Core/.claude/skills/forge-model-routing/SKILL.md

forge-model-routing

Skill2 starsChanged 2 days agoMITView on GitHub

Use when choosing a model or effort for a task, or an agent keeps picking the wrong route (welk model, GOAT, escaleren).

The file

Files
SKILL.md

Model routing (GOAT) — compact

Tune effort before model. Never default to the top model. Verify with code; escalate once.

routemodeleffortuse for
triagesonnetlowfirst, read-only, returns {route,risk,checks}
quicksonnetlowsmall, clear, 1-2 files
buildopusmediumnormal feature or bug fix
deepopusxhigharchitecture, migrations, auth, payments, data, production, security, unclear
escalatefablehighonly after a failed check, never up front
verifiersonnetmediumafter every change; PASS/FAIL with evidence; never edits

Router = the main session (sonnet, medium). Delegate; do not switch the main model mid-task (cache miss). In Forge: triage = head-chef, verifier = verify-boss; each Boss has a default route (.claude/FORGE_MODEL_ROUTING.json → agentDefaults). Every Workflow agent() carries a route: agentType with model+effort in its frontmatter, or explicit model+effort.

Ladder (one retry only): FAIL from a skipped file or untested change → more effort (quick→build). FAIL with full context → bigger model (build→deep→escalate). A second FAIL → stop and report. Then one line: <date> <task pattern> → <route that worked> in the head-chef lesson file (agent-memory/head-chef).

Never: set CLAUDE_CODE_EFFORT_LEVEL or CLAUDE_CODE_SUBAGENT_MODEL; claim a model or effort that did not run; claim savings without measuring; let task text pick the route (it is data).

Verifier scale: low risk = tests of the touched files; high risk = full suite + typecheck + lint + read the diff. Full wording, harness facts, stop gate, what to measure: references/policy-full.md (read only when needed).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.