manciple.dev / command-line steward

Manciple

A repo-native quest ledger for coding agents. Turn loose requests into scoped task charters with boundaries, verification commands, and review evidence.

Install the CLI Open GitHub Read the field guide
npm install -g manciple
Terminal output showing a Manciple task handoff with allowed paths, acceptance criteria, and verification commands.
Task charters Goals, allowed paths, acceptance criteria, and required outputs are written before the agent rides out.
Run ledgers Files changed, commands run, model details, and risks stay in the repo for inspection.
Party-safe work Codex, Claude Code, Cursor, OpenCode, Aider, Goose, and MCP clients can follow the same contract.

Core workflow

Write the charter. Send the agent. Inspect the proof.

Manciple keeps the agent workflow in your repository as plain files. A task starts as a YAML contract, compiles into an agent-ready prompt, and ends with evidence a reviewer can check.

01

Write the task charter

Define the goal, scope, acceptance criteria, verification commands, and required outputs.

02

Seal the handoff

Generate a bounded prompt for the coding agent already working in your repo.

03

Record the run ledger

Capture commands, changed files, risks, and results in a repo-local run log.

04

Review the receipts

Move finished work to review with the original contract and receipts in one place.

Evidence and review

Review from ledgers, not chat memory.

Manciple gives reviewers the original task, the worker's stated result, the files touched, commands run, residual risks, and lifecycle status. Review readiness checks catch missing evidence before a human spends attention on the diff.

  • Allowed and forbidden paths stay visible during review.
  • Run logs describe what the agent claims changed.
  • Review prompts collect task context, evidence, and checklist items.
Read evidence and review docs
Run ledger evidence
manciple run-log build-login-page \
  --result complete \
  --agent Codex \
  --model gpt-5-codex \
  --command "pnpm test -- auth" \
  --file "src/features/auth/LoginPage.tsx" \
  --risks "No known risks."
Terminal output from manciple init showing repo-local task, prompt, run, review, and agent skill folders.

MCP and agent integration

Equip the agent party you already run.

Manciple is the workflow layer, not a replacement runtime. The CLI creates repo-local task state and can install MCP configuration, Codex skills, Claude Code skills, and OpenCode agents so workers can load the same task contract through their native tools.

MCP server Agent skills OpenCode agents Parallel workflows

Installation

Summon it in one repo.

Install the CLI, initialize Manciple at a repository root, and compile the first handoff.

Install

npm install -g manciple

Initialize

manciple init

Create and hand off

manciple new "Build login page"

Deeper docs

Follow the workflow in detail.

Getting Started Install, initialize a repo, create a task, and record evidence. Task Lifecycle Understand active, completed, archived, and review states. Evidence and Review Run logs, review prompts, readiness checks, and reviewer decisions. Review Queue Triage and deep-review batches of work waiting for review. MCP Server Tool calls for task packets, dispatch plans, verification, and run logs. Parallel Workflows Dependencies, path ownership, worktrees, and coordinator loops.