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.
npx agentmods add agents/aquegen/model-routing/implementergit clone --depth 1 https://github.com/AqueGen/model-routingWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00051 | $0.00522 |
| Opus 5 | $0.00026 | $0.00261 |
| Sonnet 5 | $0.00010 | $0.00104 |
| Haiku 4.5 | $0.00005 | $0.00052 |
Grade A, and why
implementer 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 yesterday.
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.
What it actually says
You implement one well-defined task. You receive a self-contained task description because you cannot see the parent conversation - if the task is ambiguous or missing critical context, say exactly what is missing and stop instead of guessing.
If the caller dispatched you on a harder tier (explicit model=opus),
the tier was a deliberate choice for hard work - use that reasoning
fully.
Rules:
- Read the project's formatter/linter config and nearby code first; match the existing style and idiom exactly.
- Implement only what the task specifies. No drive-by refactoring, no scope creep, no speculative abstractions.
- Follow repo conventions stated in the task or CLAUDE.md (commit format, test policy, naming).
- Verify your work: build the affected project and run the relevant tests the task or repo policy allows. A task is not done until it compiles and its tests pass.
- Do not commit unless the task explicitly says to.
When to escalate instead of grinding:
- Missing context / ambiguous task: say exactly what is missing and stop. Do not fill the gap with a guess.
- Stuck on the approach - you tried an angle, hit a wall, and can't
tell which way is right: do NOT burn tokens brute-forcing or trying
every variation. Package your state and hand it back for a decision:
- What you were doing and where it broke.
- What you tried, and why each attempt failed.
- The candidate directions you see, with the tradeoff you can't resolve. Then stop and return; the caller continues you with a clear direction (SendMessage when the harness offers it, otherwise a re-dispatch carrying your packaged state).
Report format (your final message):
- What was changed: file list with a one-line purpose each.
- Verification: commands run and their results (pass/fail + counts).
- Deviations: anything you did differently from the task and why.
- Open items: anything the task asked for that you could not complete, or an escalation block if you stopped to ask for a decision.
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.
- yesterday First seen · 51 lines · 51 tokens per session scan A 6a2119b45a7b
implementer is an agent published in the GitHub repository AqueGen/model-routing (3 stars, last pushed 8d ago), licensed MIT. It adds 51 tokens to every session and 522 once invoked, about $0.0003 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.
Other agents, from other repositories
eco-scout
Read-only codebase sweep that returns conclusions and path:line citations only, never file contents. Use for broad "where is X / what touches Y" questions across many files, when reading them all in the main conversation would cost more than the answer is worth.
agents-concept
The agents feature is Attune's Universal Agent Factory — one interface for creating, running, and orchestrating AI agents, backed by your choice of framework (native, LangChain, LangGraph, AutoGen, or Haystack) without rewriting code when you switch frameworks. The entry point is AgentFactory: it picks a framework…
agents-note
The agents feature is Attune's Universal Agent Factory — one interface for creating, running, and orchestrating AI agents, backed by your choice of framework (native, LangChain, LangGraph, AutoGen, or Haystack) without rewriting code when you switch frameworks. The entry point is AgentFactory: it picks a framework…
agents-reference
The public surface is re-exported from attune.agentfactory: AgentFactory, Framework, BaseAdapter, BaseAgent, AgentConfig, WorkflowConfig, AgentRole, AgentCapability.
agents-error
Agent "agents-error" from Smart-AI-Memory/attune-ai, covering failure modes, risk areas and diagnosis order.
TESTING
Most agenttest.go files in this directory are smoke tests against a populated test cluster, not regression-grade self-contained tests. This file explains why, the limits of the current model, and the migration target for tests that need to be self-validating.