- Home
- Skills
- Agents & Orchestration
- Opus Delegate for Claude Code
More screenshots
Works with the AI Agents you already use
Opus Delegate for Claude Code
Stretch your Claude Code weekly limit: heavy work like features, refactors and long docs goes to Opus subagents automatically, while your top model keeps planning and review.
$9
Opus Delegate for Claude Code
Example session with this skill installed
Add a "forgot password" flow to my FastAPI app: reset link by email, one-time tokens that expire, rate limiting, and tests. Keep all the existing tests green and restart the server when done.
- Read your context and instructions
- Compiled the opus delegate
Summary
Real session, lightly trimmed. The orchestrator (Fable 5) turned the request into a detailed handoff brief and spawned one Opus worker. The worker built and tested the whole feature in about 3 minutes with 16 tool uses, all on Opus. The orchestrator only read the report below and reviewed it.
⏺ Agent(general-purpose, model: opus) · Build password reset flow
⎿ Done (16 tool uses · 2m 52s)
Worker report
I've added a forgot-password flow that resets by email. All tests pass, the live server was restarted, and it's committed.
What was added
| Piece | What it does |
|---|---|
| Login page | New "Forgot your password?" link |
/forgot-password | One email field with CSRF. Real and unknown emails get the same page, and the email is sent after the response, so neither the page nor the timing reveals whether an account exists. Limited to 3 requests per hour per address and 10 per IP. |
| Reset email | One line saying what to do, the link, then "If you did not ask for this, you can ignore this email. The link stops working in 2 hours." |
| Tokens | Stored hashed in a new reset_tokens table, valid for 2 hours, usable once. Using one cancels every other open link for that user. |
/reset-password | Valid link: new password and repeat fields with the same live checklist as sign-up. Bad or expired link: a friendly page with a "Send me a new link" button. |
| Sessions | Every session from before the reset stops working, through a new pw_changed_at column checked on each request. |
Decisions beyond the brief
- Reset tokens got their own table, because the existing confirmation code deletes all of a user's rows in the shared one, which would wipe reset links.
- A reset also confirms the email address, since clicking the link proves ownership.
- A reset clears the login lockout, so a locked-out user can log in straight away.
- The password checklist moved into a shared template used by both sign-up and reset.
Tests: 59 passed (the 55 existing plus 4 new), covering identical responses for real and unknown emails, the full request-to-login flow, used, expired and made-up links, weak and mismatched passwords, the old session dying, and rate limits.
Server: restarted. /, /forgot-password and /login return 200, and a bogus reset link returns its friendly 400 page.
Next steps
- Deploy as usual. Only the 8 files the worker touched were committed.
- Known trade-off: each new request issues another link without cancelling earlier ones, so a delayed first email still works. All open links die as soon as one is used.
Connects securely to your tools. The creator never sees your data.
About this skill
Claude Code meters its top tier models with tight weekly limits. If your best model types every line of code itself, the cap is gone by Wednesday. Claude Code can already spawn subagents on Opus, and Opus handles implementation work very well. Nothing makes that the default. This skill does.
What it does?
- Your orchestrating model keeps the thinking: planning, review and the final answer.
- Heavy self-contained work (features, refactors, research across many files, long docs) is handed to an Opus subagent instead of running inline.
- It routes the specialist agents you already have too, by spawning them with the Opus override. No need to edit any of their files.
- Quick two minute edits stay inline on purpose, so you never wait on a subagent for a one-liner.
What's inside
| Piece | What it does |
|---|---|
SKILL.md | The delegation rule plus how to write a good handoff prompt |
agents/opus-worker.md | A generic worker agent pinned to Opus, installed on first use |
scripts/opus.js | Optional opus terminal command: interactive Opus session, or a one-shot answer |
Real numbers from two sessions
| Session A (6 days) | Session B (10 days) | |
|---|---|---|
| Output written by the orchestrator (Fable 5.1) | 0.59M tokens (24%) | 2.8M tokens (16%) |
| Output written by Opus workers | 1.83M tokens (76%) | 14.3M tokens (81%) |
| Tasks handed to Opus | 44 | |
| Code changed | 57,147 lines added |
Across both sessions, about four of every five tokens written came from Opus. The top tier model spent its share on planning, writing handoffs and reviewing reports.
Why trust it?
No binaries, no API keys, no server, nothing running in the background. It uses documented Claude Code features (model pinning in agent files and per-agent model overrides). Every file reads in a minute, and removing it takes a minute.
Roadmap
Today the skill works in Claude Code only. The same idea, a top model that plans and reviews while a cheaper model's subagents do the heavy work, can carry over to other agents that support subagents with their own model choice. Planned next:
- Codex CLI: route heavy work from the top model to cheaper Codex subagents.
- Other agents with subagent support, such as Gemini CLI and Cursor, as their subagent features allow.
Buyers get every future version of this skill at no extra cost.
- Claude Code only. It does not work in regular Claude chat, because chat cannot start Opus subagents.
- The delegation is an instruction the model follows, not a hard router. Models follow it well, but nothing enforces it.
- Quick two minute edits stay on your main model on purpose.
- It shifts usage to Opus rather than removing it. Opus work still counts toward your all-models weekly limit.
- The worker does not see your conversation, so tasks that depend on a long chat history stay inline.
- The opus-worker agent is installed on first use and becomes available from the next session.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 19 days ago
Requires
Permissions
Write access is used once, to copy the opus-worker agent into ~/.claude/agents on first use. Shell is only used by the optional opus command, which launches the claude CLI with the Opus model. No network calls, no environment variables, no API keys.