Skip to content

Pi Agent Foreman

A Pi extension that catches the turn-ending "Continuing the run" reply and sends the agent back to it, gated on the latest user request actually containing a command to do work.

Stars
2
Language
TypeScript
License
MIT
Published by
Community
Author
alexshpunt
Last verified
Oct 2026

About this project

Your agent says "Continuing the run" and ends its turn. Foreman sends it back to do the work it left unfinished.

Quoted from the project's README. Read the full README.

Highlights

  • Two gates in series: an explicit command in the latest user request, then an announced next action or unfinished work in the final reply.
  • Six reply questions share one classifier request, and a blocker, permission need, postponement or refusal vetoes continuation.
  • If the classifier is missing, unavailable or answers invalidly, Foreman stays quiet with a warning instead of switching providers.
  • The instruction is delivered as a normal user message; the main agent gets no wrapper and no hidden transcript.

Watch out

The request gate is fixed at 0.5 and the README says lowering agentForeman.threshold does not move it, so the only tunable is the positive signal threshold. Each blocking signal vetoes at 0.5 regardless of that setting, and Foreman never combines the six scores into a single probability, which means you cannot retune the trade-off by moving one number. The agent that writes the instruction sees the final reply, the last user request and bounded activity including compact tool arguments and tool result excerpts, which the README warns can contain private data and that size limits do not redact; private thinking is not included. Back to Work is on by default and borrows your current session model until you pick one. The tuning bench makes real API calls, is excluded from npm test and CI, and the README frames a failing case as a tuning target rather than a broken build. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

22GitHub stars
Runnable agent harness that moves pick, score and yes-no loop choices to Jev or Laya in milliseconds while reserving the language model for writing, compiling DECISION.md into executable questions.
Jev Agents & MCP#community#jevbest#catalog-import
15GitHub stars
Node CLI that lints a file tree against fuzzy natural-language rules in one Markdown file, judging each matched file with a swappable TypeSafe-backed judge.
Jev Agents & MCP#community#jevbest#catalog-import
22GitHub stars
Command-line toolkit that wraps Jev judgments into scriptable verify, screen, classify, extract, find, rerank, match, route, ask, compact, and batch commands.
Jev Agents & MCP#community#jevbest#catalog-import
24GitHub stars
Wires Jev into agents as seven MCP tools, an embeddable JevDecisionModel library with run functions, and Claude Code hooks at tool and turn boundaries.
Jev Agents & MCP#community#jevbest#catalog-import
12GitHub stars
Pi extension that puts Jev behind coding-agent hooks: JSON packs ask closed questions on redacted state and turn probabilities into block, confirm, annotate, or steer actions.
Jev Agents & MCP#community#jevbest#catalog-import
7GitHub stars
MCP server exposing Jev Choice, Score, Noul, batched ask, file-reading triage, and model listing as typed tools with full probability distributions.
Jev Agents & MCP#community#jevbest#catalog-import
Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai