Intent-Router
Intent compiler skill that turns vague coding requests into a typed IntentSpec, with Jev or LAYA as optional decision backends.
Overview
Intent-Router is an agent skill that acts as an intent compiler for coding agents. Given a vague request such as add caching to the user API, it first looks up what the repository, ticket system or docs already answer, asks only the questions that genuinely need the user's judgment, and calls out requests that contradict themselves, admit two readings, or rest on a premise the repo disproves. It refuses to emit while intent is underspecified and writes a machine-readable IntentSpec to .intent/ that later sessions read before probing again. After delivery it checks the result against the spec one constraint at a time.
The README positions Jev and LAYA as the System One decision layer that assumes clear input already exists, and lists both under its backends; the default backend is prompt-only and needs nothing installed. Its own evaluation compares bare and skill-assisted OpenCode deliveries: with the skill, each run asked one question and 3 of 3 delivered caches stated a failure policy, versus none of 5 bare runs.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which reading of this request is intended?(inferred) | — |
| Yes / no | Does the delivered work satisfy this IntentSpec constraint?(inferred) | yes · no |
How it works
- User gives a vague request to the coding agent
- Skill probes repo, tickets and docs for answers
- Ask only unresolved questions; flag contradictions or ambiguity
- Emit a typed IntentSpec to .intent/
- After delivery, verify each constraint and record the result
Strengths
- Fewer questions to the user, focused on genuine judgment calls
- Spec persists across sessions and agents
- Constraint-by-constraint verification of delivered work
- Works prompt-only without any decision-model backend
Caveats
- Jev and LAYA are optional backends, not the core of the skill
- Evaluation sample is small (5 bare vs 3 skill runs)
At a glance
- Type
- Project
- Status
- beta
- Runtime
- agent skill
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Intent compiler for AI agents — converges vague requests into typed IntentSpec…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Intent-Router decide?
What a request really means, which questions still need the user, and whether delivered work meets each constraint.
Does it require Jev?
No. The default backend is prompt-only; Jev and LAYA are listed as optional decision backends.
Where is the spec stored?
In an IntentSpec YAML file under .intent/ in the repository.
Related projects
MCP server giving coding agents fast typed decisions (ship gates, risk checks, file triage, page checks) as accept, review or escalate.
MCP serverbetaTypeScriptPortable skill pack giving agents a conservative Jev second opinion on intent, completion, routing and bounded browser navigation.
ProjectbetaPythonTyped judgment layer for Claude Code, Codex and Hermes Agent: deterministic facts in code, language judgments in Jev, with receipts.
ProjectbetaPythonProduction control plane for OpenClaw agents: Jev routes each Slack message to its task, with Temporal workflows, an evaluator and memory.
LibraryproductionPythonHermes skill that turns an intention into a judgeable finish line and a run prompt, with Jev deciding fuzzy requirements at the gate.
Agent skillbetaPythonChinese-language agent skill that offloads classify, screen, score and compliance checks to Jev via OpenCode Zen's free tier.
PluginbetaPython