pi-jev-todo-audit
Pi extension that audits whether the rpiv-todo board matches visible work and nudges the agent when it drifts off-task.
Overview
pi-jev-todo-audit is a Pi extension that checks whether the @juicesharp/rpiv-todo board matches the agent's visible work. Per the list summary, it audits board drift every 10 agent loops and injects corrective nudges when the agent wanders off-task. It sends finite, task-specific change questions to the pi-llm-as-jev judgment service, which selects a native classifier or LLM backend.
The judge acts as a macro-level engineering lead: it sees goals, task definitions, reported outcomes, progress, user decisions and open questions, but not execution detail. The extension is advisory: it replays persisted todo snapshots, never mutates tasks directly, and asks the main agent to make justified updates. Audit state lives in non-context session entries, with no summariser model or generated free-text memory. On first load it installs and registers the judgment service package through Pi's package manager if none is selected, respecting existing pins and scopes.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Has this task's status changed according to visible work?(inferred) | yes · no |
| Yes / no | Is the agent working off the todo board?(inferred) | yes · no |
How it works
- Every N loops, gather board state and visible progress.
- Ask the judgment service finite change questions.
- Compute safe advice from answers.
- Nudge the main agent to update the board.
Strengths
- Advisory; never edits tasks directly.
- No generated memory or summaries.
- Incremental protocol with persisted cursors.
Caveats
- Depends on rpiv-todo and pi-llm-as-jev.
- Auto-installs a service package on first load.
Getting started
pi install git:github.com/xz-dev/pi-jev-todo-auditAt a glance
- Type
- Plugin
- Status
- beta
- Language
- TypeScript
- Runtime
- Pi extension
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Pi extension that audits rpiv-todo board drift every 10 agent loops using TypeSafe's jev model, injecting corrective nudges when the agent wanders off-task.
hellogumbo/awesome-jev
Frequently asked questions
Does it change my todo list?
No, it only advises the agent to update it.
Which judge does it use?
The pi-llm-as-jev service, which can use a native classifier or LLM backend.
How often does it audit?
Every 10 agent loops, per the list summary.
Related projects
Pi extension that sends agents back to work when they stop early, using six Jev questions to decide if continuation is warranted.
ProjectbetaTypeScriptMulti-agent Codex factory: Jev decides task shape, decomposition and routing, and failures are classified, not blindly retried.
ProjectprototypePythonSoftware-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonOh My Pi coding agent (a Pi fork with IDE wiring) with an optional TypeSafe judgment provider for bounded decisions.
ProjectproductionRustMCP server giving coding agents fast typed decisions (ship gates, risk checks, file triage, page checks) as accept, review or escalate.
MCP serverbetaTypeScriptAgent plugin that asks Jev whether the session is at a safe boundary to /compact and can trigger it on Pi and Claude Code.
PluginbetaTypeScript