PLATE

PLATE (Process Lifecycle Agentic Task Engine) is a GitHub-native methodology and tooling monorepo for agent-driven software delivery. Humans keep judgment; agents do the toil; GitHub preserves truth.

This repository (akasper/plate) is the implementation monorepo. The installable Python package on PyPI is still named plate-core (import path plate_core); the product name is PLATE.

Surface	Target User	How to Install
gh plate extension	Humans and scripts — terminal PLATE operations	gh extension install akasper/gh-plate
plate-mcp MCP server	AI agents — first-class tool calls via MCP	pip install plate-core then plate-mcp (or python -m plate_core.mcp_server)
CLI agent plugin (Copilot CLI, Grok Build, …)	Interactive sessions — plate persona + MCP wiring	pip install plate-core then marketplace install from this repo (see below)

All surfaces share the same library code under src/plate_core/, so CLI, MCP, and plugins stay in parity.

Licensing

This project is licensed under a source-available model (MIT base + Commons Clause License Condition v1.0). Free for non-commercial/personal/internal use and modification. Commercial use, resale, or SaaS offerings require a separate license. See the LICENSE file for the full text.

What It Does

PLATE tooling reads GitHub + local .plate / .agentic/ state and drives process loops:

• Health & routing — gh plate health, gh plate what-next / plate_what_next (priority ladder: budget, open PRs, adoption, self-migrate, ready work, PM, epics)
• Autonomy — budgeted AutonomyEngine (.plate autonomy.risk_tolerance), procedures, checkpoints, ledger, shadow/simulate for high-impact actions
• Project Manager — long-running orchestrator (gh plate pm, fleet handoffs, loop ticks); browser UI deferred
• Endless feed + Q&A planning — Questions/Tasks feed; product/feature/release planning sessions with approval
• Adoption & self-migrate — under-30m local path (gh plate adopt, import-payload, session timer); pin/payload self-migrate plan + offline verify
• Release ceremony — multi-track release status, fragments, cut/finalize helpers
• PR green loop — gh plate pr babysit with feedback resolution and base sync strategies
• Epic / feature / bug loops — status, stubs, feature/bug stage machines, media + design contracts
• Baseline agents & skills — catalog via gh plate agents / gh plate skills
• E2E Playwright helpers — scaffold/validate/record via MCP
• CLI agent plugin — default plate persona when PLATE signals are present (AGENTS.md, .plate/)

See AGENTS.md, SPEC.md, and docs/wiki/V1-Autonomy-Surfaces-Epic-Closeouts.md for operating rules and first-slice surface status. Release #654 tracks v1.0.0 readiness (do not claim 1.0 without checklist E2E).

Quick Start

As a Python package (plate-core on PyPI + MCP)

pip install plate-core
plate-mcp   # stdio MCP server for agents
python -c "import plate_core; print(plate_core.__version__)"

As a gh extension

gh extension install akasper/gh-plate
gh plate health --json
gh plate what-next --json
gh plate autonomy --status
gh plate pm --status
gh plate feed --json
gh plate release status
gh plate adopt --json
gh plate self-migrate --verify --json
gh plate epic status --repo akasper/plate --json
gh plate bootstrap --repo OWNER/REPO --adopt --json
gh plate pr babysit 112 --repo akasper/plate --json

The gh plate extension is published from a dedicated thin repository (akasper/gh-plate) so the GitHub CLI name starts with gh-. Implementation, the plate-core package, MCP entrypoint, plugins, and source live in this monorepo (akasper/plate).

As an MCP server (v1 baseline; works in Copilot CLI, Grok Build, and other compatible agents)

# In a supported CLI agent session (e.g. Copilot CLI):
/mcp connect /absolute/path/to/plate/plate-mcp
# Then call tools: plate_health, plate_epic_status, plate_features, plate_bootstrap, plate_plan_epic, plate_pr_babysit, plate_resolve_review_thread, plate_agents, plate_agent, plate_skills, plate_skill

As a CLI agent plugin (Copilot CLI, Grok Build, and other standards-compliant CLIs)

# Install the runtime prerequisite first so the plugin's MCP command is available.
pip install plate-core

# Register this repository as a marketplace, then install the plugin from it.
copilot plugin marketplace add akasper/plate
copilot plugin install plate-core@plate-marketplace

# Grok Build (TUI / CLI):
grok plugin marketplace add akasper/plate
grok plugin install plate-core@plate-marketplace --trust
# Then enable the plugin, reload the TUI (`r` in Plugins), and verify:
#   grok inspect   # expect plate-core skills, plate agent, and plate-core (stdio) MCP
# Or reinstall from a local checkout during development:
#   grok plugin install /absolute/path/to/plate/.plugin --trust

# In a new session with your CLI agent, invoke the plate agent (see your agent's docs for the exact command, e.g. /agent plate)

For local development or direct-source installation, these equivalent commands also work (adjust for your CLI):

copilot plugin install /absolute/path/to/plate
# or
copilot plugin install akasper/plate:plugin
# Grok equivalent: grok plugin install /absolute/path/to/plate --trust

The marketplace flow is the supported public install path. The plugin still expects the plate-mcp command to be available on PATH, which is why pip install plate-core remains a prerequisite until publication/runtime provisioning is further automated. There is no separate GitHub-run submission process for Copilot CLI or Grok Build marketplaces: this repository itself becomes the marketplace once the manifest is merged to the default branch and you treat that path as the supported public install channel.

Marketplace release checklist

Full maintainer checklist (surfaces, smoke, human Tasks): docs/bootstrap/marketplace-install-checklist.md (#378 / #379).

Before cutting a release that ships marketplace install:

1. Confirm .github/plugin/marketplace.json (Copilot → source: "plugin") and .grok-plugin/marketplace.json (Grok → ./.plugin) still point at the intended plugin payloads; versions match plate-core.
2. If baseline catalog skills changed, run python3 scripts/generate-plugin-skills.py (and commit) so plugin/SKILLS.md, plugin/skills/*/SKILL.md, and the mirrored .plugin/ copies stay in sync; then python3 scripts/generate-plugin-skills.py --check.
3. Re-run python3 scripts/generate-grok-plugin-index.py (and commit) if plugin/agents/, plugin/skills/, .mcp.json, or manifest keys changed; then python3 scripts/generate-grok-plugin-index.py --check.
4. Verify the runtime prerequisite is available with pip install plate-core (plate-mcp on PATH).
5. Smoke-test the pre-launch install flows (Copilot + generator for Grok):

   copilot plugin marketplace add akasper/plate
   copilot plugin install plate-core@plate-marketplace
   python3 scripts/generate-plugin-skills.py --check
   python3 scripts/generate-grok-plugin-index.py --check
   pytest tests/test_copilot_cli_marketplace_packaging.py -q

6. Complete the human-owned publication tasks tracked in #380 and #381 (and #625/#626 when PyPI pins require them). Agents must not complete those Tasks.
7. Fold marketplace work into the active Next Release issue (#612) and cut through the normal PLATE release ceremony.

See the grok-build epic for full CLI-agnostic details and verification that no vendor-specific language remains in the plugin files. (This release also closes the Grok marketplace discovery gap reported in #570.)

Playwright E2E Testing

PLATE includes tools for scaffolding, validating, and managing Playwright E2E tests:

MCP Tools

• init_playwright — Initialize Playwright E2E setup in a repository

  # Copy config, test specs, and recording scripts from plate's template payload
  @copilot init-playwright repo_path="/path/to/repo"

• validate_e2e_tests — Verify Playwright setup and detect missing configuration

  @copilot validate-e2e-tests repo_path="/path/to/repo"

• record_e2e_gif — Record and generate demo GIF from a Playwright E2E test

  @copilot record-e2e-gif repo_path="/path/to/repo" test_name="feature-name" quality="medium"

CLI Feature Detection

Check if a repo has Playwright E2E setup:

gh plate features --repo owner/repo

Output example:

Repo: akasper/plate_template

Autonomous Mode.................... ✅ ENABLED
Platform Monitor Workflow.......... ⏹️  NOT CONFIGURED
Copilot Plugin (.plugin)........... ✅ ENABLED
Copilot Plugin (plugin)............ ✅ ENABLED
MCP Manifest (.plugin)............. ✅ ENABLED
MCP Manifest (plugin).............. ✅ ENABLED
Per-feature change files........... ✅ ENABLED
Release notes..................... ✅ ENABLED
Baseline Agents Catalog........... ✅ ENABLED
Playwright E2E Testing............. ✅ ENABLED

Runtime layout (v1 baseline)

plate/                     # monorepo (product: PLATE)
├── AGENTS.md              # operating rules (source of truth with GitHub)
├── SPEC.md                # product intent
├── .plate                 # local autonomy/release config (JSON)
├── .agentic/              # fragments, procedures, costs, PM queue, …
├── .plugin/               # root plugin discovery + MCP config
├── .github/plugin/        # Copilot CLI marketplace manifest
├── plugin/                # plugin source (agents, skills)
├── src/plate_core/        # Python package (PyPI name: plate-core)
│   ├── health.py, what_next.py, autonomy.py, pm.py, feed.py, …
│   ├── adoption.py, self_migrate.py, release.py, pr_babysit.py, …
│   ├── cli.py, mcp_server.py
│   └── data/baseline_catalog.yml
├── docs/wiki/             # durable epic closeouts, Goals, …
├── gh-plate               # gh extension entrypoint
└── plate-mcp              # MCP server entrypoint

Contributing

This repository follows the PLATE methodology. See AGENTS.md for agent operating rules and the full PLATE workflow.

---

Keeping Your Fork Current

If your repository started from an older plate_template release and has local process customizations, avoid full-file replacement during upgrades.

Use this sync flow:

1. Fetch upstream template updates (git fetch upstream) and review diffs for AGENTS.md, .agentic/skills.yml, .agentic/releases/, and workflows in .github/workflows/ that contain PLATES-CORE markers.
2. Import only upstream-owned PLATES-CORE sections into your customized files.
3. Preserve local sections outside those markers.
4. Open an atomic PR with the correct PR type label and issue linkage (Closes #N when applicable).
5. Update per-feature change files in .agentic/releases/ with imported behavior and evidence.
6. Run required checks before merge.

This keeps downstream repos aligned with new core PLATE behavior without erasing project-specific policy.