pi-jev-code
Single-agent Jev coprocessor for Pi: semantic pre-edit gates, evidence checks, diff review and append-only telemetry.
Overview
pi-jev-code is a Jev coprocessor for Pi coding sessions in which your normal Pi model stays the only coding agent, with no planners, subagents or orchestration. Jev provides semantic gates around the work: a pre-edit judgment that can block an edit for scope, target, design, requirement or evidence concerns; a jev_evidence tool the agent can use to support a premise; ranking of search results; and a post-turn diff review from baseline to current that returns PASS, REVIEW or FAIL, where review or fail can trigger a same-agent follow-up.
Version 0.5 added append-only observability telemetry kept out of the model context, and 0.5.1 calibrated behaviour from a real session: strong explicit evidence can resolve an evidence-only block, block outcomes are tracked, deterministic installs and formatters advance the edit version without a Jev call, and search ranking only pays for real paths. Version 0.5.2 fixes a runtime regression.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is there insufficient evidence or a scope/design concern for this edit? | yes · no |
| Choice | Post-turn diff verdict | PASS · REVIEW · FAIL |
| Yes / no | Does the evidence support this premise? | supported · not supported |
How it works
- Before an edit, ask Jev pre-edit questions; block on concerns.
- Agent can gather evidence with jev_evidence to resolve blocks.
- After the turn, review the diff from baseline.
- Log telemetry for every judgment.
Strengths
- Keeps a single coding agent.
- Measurable block usefulness via telemetry.
- Calibrated from real sessions.
Caveats
- Pre-edit blocks can add friction.
- Requires a TypeSafe key.
Getting started
Install the pi-jev-code extension into Pi and set your TypeSafe key.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Pi extension
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Single-agent Pi coding coprocessor with Jev semantic gates, baseline-to-current diff review, and append-only observability telemetry.
hellogumbo/awesome-jev
Frequently asked questions
Does it add extra agents?
No, Pi's model remains the only coding agent.
What happens after a turn?
Jev reviews the diff and returns PASS, REVIEW or FAIL.
Is telemetry sent to the model?
No, it is append-only and kept out of the model context.
Related projects
Desktop app running coding agents in parallel git worktrees, with optional Jev judgments of each changed file against the task.
AppbetaSoftware-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonJev-first CLI coding agent that routes a request to a deterministic workflow which asks Jev fixed-choice questions about bounded evidence.
ProjectprototypeTypeScriptMCP server giving coding agents fast typed decisions (ship gates, risk checks, file triage, page checks) as accept, review or escalate.
MCP serverbetaTypeScriptLocal desktop pal that verifies what a coding agent really did, with optional Laya or Jev double-checks of unclear test results.
ProjectbetaTypeScriptDeepSeek Harness plugin checking whether an agent's 'done' claim is true, using deterministic evidence plus one batched Jev assessment.
PluginbetaTypeScript