defer

A model-routing plugin that decides where a piece of work should run and produces the command to send it there. It uses task categories, capability results, model effort settings, and available budget information.

In plain words
What is it for?
Use it to route coding or other delegated tasks to a suitable model, generate the required command, and check the routing result.
Why use it?
It removes repeated guesswork about which model to use and how to verify that the selected route actually ran.

Plugin for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Claude Code
/plugin marketplace add fledgeling-co/fledgeling-plugins
agentmods
npx agentmods add plugins/fledgeling-co/fledgeling-plugins/defer
Clone the repo
git clone --depth 1 https://github.com/fledgeling-co/fledgeling-plugins

Made for: Claude Code.

Per session not measured What this adds to a session before it is invoked.
When invoked not measured Not applicable: nothing here is loaded into a session.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Security

Grade A, and why

defer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

plugins/defer/.claude-plugin/plugin.json · 10 lines

What it actually says

{
  "name": "defer",
  "description": "Decide which model a piece of work goes to, and produce the exact command that sends it there. One routing policy for every skill that hands work outside the session: six task classes mapped to ten lanes across five model families, each pinned to a model and an effort, with the CLI arguments, the environment GLM needs, and the check that proves the lane ran as routed. Narrows those lanes by what the work actually is, against a capability matrix measured over 106 benchmark tasks, so a piece of work goes to the cheapest lane that scored level with Opus on that shape rather than to Opus by default. Where several lanes remain it picks the one with the most plan headroom per remaining day, measured — Claude and Codex report a real utilization percentage that is already on disk, and Grok, GLM and Gemini are counted locally against a budget calibrated from a percentage you can see. Ships a lane prober, a policy selftest that runs no model, and a wire-verify method per lane, because the most expensive failure in a multi-model system is a plausible answer from the wrong one.",
  "version": "1.4.0",
  "author": {
    "name": "Fledgeling",
    "email": "[email protected]"
  }
}
Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 10 lines scan A 6bc984a1079c

Subscribe to this mod's changes

defer is a plugin published in the GitHub repository fledgeling-co/fledgeling-plugins (2 stars, last pushed 2d ago), licensed MIT. Its token cost is not measured: this kind of file is read by the harness, not the model. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other plugins, from other repositories

supergoal

Plan and autonomously build a software task end-to-end. Recons the codebase, applies preloaded memory, decomposes into the right number of phases, gets one confirmation, then prepares a single ready-to-paste /goal command — one paste between you and done — that drives execution to completion with built-in retry…

robzilla1738/supergoal · not measured

context-graph-plugins marketplace

Runtime plugins for Agent Context Graph.

memgraph/ai-toolkit · not measured

use-claude-limits-efficiently

Budget loop for long-running or parallel agent work under Claude's 5-hour and weekly usage limits — run bounded waves, check real usage between them, pause before the cap, and resume with self-contained prompts only when the window is actually clear.

patrickdappollonio/claude-plugins · not measured

coaltipple

CoalTipple — model/effort router for Claude Code. Delegates a task you CAN do but is large+cheap DOWN to a lower tier to save tokens; escalates a task beyond the current tier UP for quality — up to Fable 5, the top rung (a consent-gated, real-money spawn). Two-knob routing (tier × effort, effort-before-tier), an alias.

TheColliery/CoalTipple · not measured

gemini-swarm

Make Gemini your default writer and image generator: the orchestrator plans and reviews while parallel agy (Antigravity CLI / Gemini) sessions do the writing, plus a gemini-imagegen skill that generates real images via Gemini's generateimage tool.

Vallykrie/gemini-swarm-skill · not measured

harness-maker

Per-project AI coding harness for Claude Code · Cursor · Codex. Profiler + 10-dim interview build a different harness for every project.

Ecro/harness-maker · not measured