light-coder-orchestration

A workflow for assigning bounded coding tasks to lower-cost AI models and having another model verify their work. A bounded task has a clearly limited scope, files, and acceptance checks.

In plain words
What is it for?
Use it to split larger coding projects into independent work packages, run them through budget models, verify the results, and commit completed work.
Why use it?
It lets lighter models handle focused implementation while a separate verification step checks types, tests, and the resulting changes for mistakes.

Skill for Claude CodeCodex

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 skills/agentproto/ts/light-coder-orchestration
Any agent
npx skills add agentproto/ts --skill light-coder-orchestration
Clone the repo
git clone --depth 1 https://github.com/agentproto/ts

Made for: Claude Code, Codex.

Per session 165 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,838 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. 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 $0.00165 $0.01838
Opus 5 $0.00082 $0.00919
Sonnet 5 $0.00033 $0.00368
Haiku 4.5 $0.00016 $0.00184

Measured 2d ago against content hash e5d2acb8f062, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

light-coder-orchestration 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.

packages/skill-pack-agentproto/src/skills/light-coder-orchestration/SKILL.md · 143 lines

How it starts

The opening of the file, as written. The whole thing — 143 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Light-coder orchestration (budget models + Sonnet safety net)

Methodology proven on a real project (monorepo extraction, ~10 WPs delivered green). The orchestrator (you, in cowork) does not code: it breaks work into bounded WPs, has a light model execute via agentproto, and has Sonnet verify. Light models are good at bounded implementation but have blind spots → the Sonnet verification is not optional.

One-line principle

bounded brief → execution (light model) → green gate (check-types/tests) → Sonnet verification pass → commit

1. Wiring up a light model (hermes via OpenRouter)

  • Adapters via adapter_list. claude-code and hermes are ACP.
  • hermes routes lots of OpenRouter models (glm-5.2, deepseek-v4-pro, kimi, qwen, grok, gpt-5.x…). Selection: agent_start(adapter:"hermes") then send /model <id> as the first prompt/model works in hermes ACP (≠ claude-code, where /model is blocked and there is no model param at spawn). Check the output: Model switched to: … · Provider: openrouter.
  • Daemon prerequisite: the provider key must be in the daemon's environment (e.g. OPENROUTER_API_KEY) — hermes acp inherits the daemon's env. Symptom when missing: "No LLM provider configured" on /model. (A selection made in the interactive hermes TUI is NOT inherited by ACP sessions.)
  • grok-4.3 often works out of the box (Nous OAuth, file ~/.hermes/auth.json); the other models go through their env key.

2. The brief (bounded, self-contained)

A good WP brief:

  • Explicit file scope + clear no-go zones ("do not touch X").
  • Recent context the model cannot guess (recent migrations, API renames…).
  • Gate: the exact commands that must be green (check-types, test).
  • STOP-if-fork: "if a non-trivial design choice comes up, STOP and ask" + name the likely forks and the preferred default.
  • Ask for a final report: files touched, design choices, exit codes.
  • Before delegating, paste the Brief Contract from supervisor-session into each brief.

Read the full file on GitHub · 143 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 · 143 lines · 165 tokens per session scan A e5d2acb8f062

Subscribe to this mod's changes

light-coder-orchestration is a skill published in the GitHub repository agentproto/ts (5 stars, last pushed 2d ago), licensed Apache-2.0. It adds 165 tokens to every session and 1,838 once invoked, about $0.0008 per session on Opus 5. 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 skills, from other repositories

strands-review

Local preview of the strands-agents/devtools /strands review agent. Body is the upstream Task Reviewer SOP verbatim — do not paraphrase. Use when the user types /strands-review, asks for a "strands review" of a PR, or wants to anticipate what the remote /strands review GitHub Action will flag. Findings are close but…

strands-agents/harness-sdk · 125 tokens

docs-writer

Draft or rewrite Strands Agents documentation pages. Use when writing new doc pages, rewriting pages that failed audit, drafting sections for existing pages, or writing blog posts and release notes about Strands. Also triggers on "write a doc", "draft a page", "rewrite the quickstart", "add a tutorial for X"…

strands-agents/harness-sdk · 74 tokens

pr-create

Creates a GitHub pull request using the gh CLI. Use when the user asks to create, open, or submit a PR on GitHub.

strands-agents/harness-sdk · 32 tokens

document-service

This skill should be used when the user asks to "analyze this codebase", "document this service", "generate technical docs", "I inherited this code", "help me understand this system", "create docs for this project", "what does this system look like", "onboard me to this codebase", "this codebase has no docs"…

awslabs/agent-plugins · 141 tokens

aws-step-functions

Build workflows with AWS Step Functions state machines using the JSONata query language. Covers Amazon States Language (ASL) structure, state types, variables, data transformation, error handling, AWS service integration, and migrating from the JSONPath to the JSONata query language.

awslabs/agent-plugins · 57 tokens

hyperpod-version-checker

Check and compare software component versions on SageMaker HyperPod cluster nodes - NVIDIA drivers, CUDA toolkit, cuDNN, NCCL, EFA, AWS OFI NCCL, GDRCopy, MPI, Neuron SDK (Trainium/Inferentia), Python, and PyTorch. Use when checking component versions, verifying CUDA/driver compatibility, detecting version mismatches…

awslabs/agent-plugins · 120 tokens