- Home
- Skills
- Agents & Orchestration
- Subagent Triage
Works with the AI Agents you already use
Subagent Triage
Optimizes subagent costs and performance by routing tasks to the right Claude model and reasoning level.
$12
Subagent Triage
Example session with this skill installed
Start a subagent to scan the /src directory and list all files that don't have a corresponding test file in /tests.
- Read your context and instructions
- Compiled the subagent triage
I have analyzed the subtask and determined the optimal resource allocation using the TypeSafe Jev model. The task is categorized as mechanical/local, requiring low reasoning effort and posing zero operational risk.
Subagent triage: haiku / low. tier fast (0.88) · effort 0.08 (0.94) · risky 0.01
Subagent Configuration
| Parameter | Selected Value | Rationale |
|---|---|---|
| Model | haiku | Task is a simple file system inventory; high confidence in fast tier. |
| Effort | low | Minimal reasoning required for existence checks. |
| Risk | low | Read-only operation; no production impact. |
Next steps
- Initiating the subagent with the
haikumodel to perform the directory scan. - Reviewing the list of missing test files returned by the subagent.
- Deciding if a subsequent high-effort subagent is needed to generate the missing tests.
Connects securely to your tools. The creator never sees your data.
About this skill
Standard agent delegation often wastes money by using high-end models for trivial tasks or fails by starving complex tasks of reasoning power. This skill implements an automated routing layer between your main agent and its subagents. It determines the optimal model tier and reasoning effort for every subtask before execution.
What it does
- Dynamic model selection maps subtasks to Haiku, Sonnet, or Opus based on task complexity.
- Reasoning effort tuning adjusts reasoning levels from low to extra-high based on the cognitive load required.
- Risk assessment automatically forces high-end models for production changes or high-stakes data operations.
- Calibrated routing uses confidence thresholds to prevent down-sampling tasks without high certainty.
- Customizable ladders allows developers to define their own model hierarchy and cost constraints in local config files.
How it works
- Analyze subtasks by generating a concise title and description of the work without leaking sensitive data.
- Query a decision model (TypeSafe's Jev or Google's Gemini Decision) to score task difficulty, reasoning needs, and operational risk.
- Apply decision logic to compare the subtask requirements against the user's defined model ladder and confidence bars.
- Invoke subagents with specific
modelandeffortparameters for the Claude Code Agent tool.
Frameworks & tools
Built for Claude Code. Uses a decision model via REST API: TypeSafe's Jev or Google's Gemini Decision (gemini-decision-1), whichever key you set. Jev is preferred when both are set. Supports local configuration through CLAUDE.md.
Why this beats prompting it yourself
Manual prompting for subagent routing is inconsistent and often ignored during deep task execution. This skill uses a dedicated decision model with calibrated confidence scores, ensuring cost savings and performance reliability are enforced mathematically rather than through vibes.
Use cases
- Cost optimization for large-scale code refactors where many subtasks are simple renames or documentation updates.
- Security shielding for infrastructure tasks that require the highest reasoning capacity to avoid destructive errors.
- Automated QA where test execution needs low effort but test architecture needs high effort.
Known limitations
Currently optimized for Claude Code. Support for Codex, OpenCode, and Cursor is limited to retrieving decisions without automated application. Requires a TypeSafe or Gemini API key.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Optimized for Claude Code. Requires TYPESAFE_API_KEY. Other IDE agents like Cursor can see decisions but not yet apply them automatically. Data privacy checks skip routing for confidential info.
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 buying a skill.
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 5 days ago
- Accounts
Google AI
Typesafe
You need one API key, not both: TypeSafe (Jev) or Google AI (Gemini Decision). Set it as TYPESAFE_API_KEY or GEMINI_API_KEY.
Permissions
Allowed Hosts