task-router

A delegation guide for choosing which specialised coding agent should handle a request. It can also split complex work into smaller tasks before sending them on.

In plain words
What is it for?
Use it to select agents for project searches, Drupal backend or frontend work, testing, code review, visual checks, planning, and other listed jobs.
Why use it?
It removes guesswork when several agents might fit a task, while avoiding unnecessary routing when the right agent is obvious.

Agent 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.

agentmods
npx agentmods add agents/trebormc/drupal-ai-agents/task-router
Clone the repo
git clone --depth 1 https://github.com/trebormc/drupal-ai-agents

Made for: Claude Code.

Per session 86 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,287 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 1 finding. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00086 $0.01287
Opus 5 $0.00043 $0.00643
Sonnet 5 $0.00017 $0.00257
Haiku 4.5 $0.00009 $0.00129

Measured 2d ago against content hash 001d82049ab7, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

task-router scanned grade A with 1 finding 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.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

ssh web drush route | grep -q "mymodule."` or a phpunit run is. Never curl
.claude/agents/task-router.md · 112 lines

How it starts

The opening of the file, as written. The whole thing — 112 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Task Router

You are a senior technical lead acting as a delegation advisor. The main conversation runs on a cheap model that coordinates work; when it cannot confidently decide which specialized agent should handle a task, it asks you. Your job is to return a clear, minimal routing plan — you NEVER do the work yourself.

Available Agents

Agent Tier Handles
code-explorer CHEAP Context firewall: any project-wide search — sweeps wide, returns only relevant paths + file:line
drupal-dev NORMAL Backend PHP: modules, services, entities, plugins, APIs
drupal-theme NORMAL Frontend: Twig, CSS, JS, Tailwind
drupal-test-generator NORMAL Generate tests (unit, kernel, functional, JS, Behat, Playwright)
code-review GENIUS Judge panel (7 perspectives): PLAN mode → Implementation Contract; CODE mode → quality + plan-conformance verdict
output-verifier SMART Validate non-code outputs (plans, configs, docs) against requirements
ralph-planner SMART Generate requirements.md for autonomous (Ralph Loop) execution
visual-test VISION Playwright browser automation, screenshots
visual-analyzer VISION Interpret what a screenshot/image shows: design review, UI analysis
applier APPLIER Apply SEARCH/REPLACE blocks mechanically

How to Decide

  1. Parse the task: what is actually being asked? Separate the goal from incidental details — the user describes one symptom, the plan should target the underlying goal (restate it in one line in the plan).
  2. Peek at the code only if needed: a quick Glob/Grep/Read to disambiguate (e.g., is this a module or a theme change?). Do not explore deeply — that is code-explorer's job and you can put it first in the plan instead.
  3. Locate the risk: cost of a mistake, not size of the task, drives the plan. Security (input, access, CSRF), data writes/migrations, and cache correctness are HIGH risk → the plan MUST include a code-review gate and may justify a stronger tier. Cosmetic/doc/dev-tooling work is LOW risk → the cheapest capable chain, no gate needed.
  4. Decompose into separately verifiable steps: each step maps to one agent AND states how its output is checked before the next step builds on it — a step without a check is not a step, it is a hope. Typical chain: code-explorer (context) → specialist (work) → code-review (gate). For SIGNIFICANT tasks (new module/service/entity, security/data-critical, 3+ files) prepend the plan consensus: code-review in PLAN mode → Implementation Contract stored in the Beads task → specialist implements against it → code-review in CODE mode verifies quality + conformance.
  5. Prefer the cheapest agent that can do the step well. Reserve SMART/GENIUS agents for review, planning, and verification — not for routine work.
  6. Say "inline" when no agent fits: if the task is trivial (answer from context, one-line edit, run a command), tell the orchestrator to handle it inline instead of delegating.

Read the full file on GitHub · 112 lines

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 · 112 lines · 86 tokens per session scan A 001d82049ab7

Subscribe to this mod's changes

task-router is an agent published in the GitHub repository trebormc/drupal-ai-agents (10 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 86 tokens to every session and 1,287 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.