Route agent tasks by capability, risk, and context budget before loading files.
Agent Routing Kit is a deterministic routing helper for coding-agent workflows. It scores a task against a small capability catalog, marks public/private risk, and estimates context usage before a session spends tokens on the wrong lane.
Large workspaces need explicit routing. A route summary and handoff receipt give an agent enough shape to start in the right place without pulling private assumptions or unbounded context into a public task.
python -m pip install -e .
agent-route "write tests and docs for a small Python CLI" --json
python -m pytest- Route a plain coding task and inspect the selected capability.
- Route a task with
--include-privateand compare omitted risk counts. - Import
route_taskandestimate_contextfrom Python.
Public-safe Python package, CLI, and Codex skill extraction. It is advisory routing, not an authorization or policy engine.
Deterministic task-routing and context-budget helpers -- explicit edges, no deps.
Agent Routing Kit is a tiny routing and context-budget utility for agentic development work. It makes a task's likely lane visible before a model session starts loading files, spending context, or crossing from public utility work into private workflow assumptions.
This is the public-safe extraction of a larger private workspace pattern: make capability routing explicit, keep context budgets visible, and separate publishable utility code from private operational systems.
- Scores a task against a small capability catalog.
- Labels capabilities as
public,private, orrestricted. - Excludes private/restricted lanes by default and reports omitted risk counts.
- Estimates context usage with a conservative token approximation.
- Emits concise Markdown or JSON handoff summaries.
- Ships as both a Python package and a Codex plugin skill.
It does not include private agent corpora, credentials, target data, live operations, exploit logic, broker integrations, or proprietary project state.
python -m pip install -e .agent-route "write tests and docs for a small Python CLI" --json
agent-route "design a frontend dashboard" --include-privatefrom agent_routing_kit import route_task, estimate_context
result = route_task("write docs and tests for a Python package")
budget = estimate_context("some model-bound context", max_tokens=8000)
print(result.to_markdown())
print(budget.status)See USAGE.md for an install line, the full CLI flag set, the
importable API, and worked examples with their expected output. A runnable
end-to-end script lives in examples/demo.py.
The repo root is also a Codex plugin. The plugin provides a compact skill for:
- choosing a simple route for a coding task,
- splitting harmless utilities from larger private systems,
- checking context budget before loading files,
- preparing public-safe handoff summaries.
- Prefer deterministic scoring over hidden model state.
- Keep private and restricted lanes out of default output.
- Show when private or restricted matches were omitted.
- Treat routing as advisory, not authority.
- Keep small tools independent enough to publish.
- Make claims testable.
- Zain Dana Harper - author and maintainer.
- Claude Code - AI coding contributor for implementation and packaging assistance.
The maintainer reviews and owns public release decisions, security posture, and final claims.
Zain Dana Harper -- small tools with explicit edges. Portfolio · HarperZ9 Built with Claude Code; reviewed, tested, and owned by me.
Keep the public README, package metadata, and examples aligned with current behavior. Before opening a PR or pushing a release, run the local package verification path.
python -m pip install -e ".[test]"
python -m pytestZentropy Labs · order out of entropy. An independent lab building evidence-first tools that leave a re-checkable artifact behind. Built by Zain Dana Harper in Seattle. The full workbench is at Project Telos.