Cloud Code Plugins/add-project-management

add-project-management

Cross-actor project management for any agent - GitHub Issues as the single source of truth, uniform task anatomy with an approval-ready completion lattice, an autonomous project steward, and projection sync. Works standalone; no dependency on fleet infrastructure.

Moved into agent-dev. The standalone add-project-management plugin is deprecated - the marketplace keeps a pointer stub for one release so existing installs don't break, but it installs nothing. The skill now lives in the agent-dev plugin as /agent-dev:add-project-management. Nothing about the standard changed - same PROJECT_STANDARD.md, same five runtime skills; only the command moved.

Installation

/plugin install agent-dev@abilityai

# Then, inside the agent that should manage projects:
/agent-dev:add-project-management

It is a single installer skill. Running it writes the standard and the runtime skills directly into the agent - there is no plugin dependency at runtime. It requires a GitHub repository and an authenticated ghCLI, and asks four configuration questions: the registry repo, the operator (the human it escalates to), the managing agent's name, and how long a task may sit unverified before escalating.

What Gets Installed

Artifact	Purpose
PROJECT_STANDARD.md	Convention doc at the repo root - the deployer's config surface. All runtime skills read it; edit this file to change behavior, not the skills.
/project-init	Create or adopt a project: GitHub epic + idempotent labels + workspace stub
/project-task	The only sanctioned interactive task-creation path - enforces full task anatomy; supports --headless for cron/compose use
/project-steward	Autonomous sweep: verify pending-verification claims, dispatch, escalate, digest
/project-reconcile	Projection sync - Google Tasks adapter v1 plus an adapter contract for other surfaces
/project-intake	Headless intake primitive: route actionable items from any source into the registry, dedupe by meaning, return the issue number - called by other skills and crons, never interactive

The Standard

Humans, the managing agent, and other fleet agents collaborate through a shared vocabulary of labels, task anatomy, and a completion lattice - all recorded in GitHub Issues:

•One registry, write-authoritative - GitHub Issues are the sole authoritative record: one epic issue per project, one task issue per task. Projections are read-only views and never write state back.

•Approval-ready completion lattice - open → pending-verification → done. An agent claiming completion moves a task to pending-verification; the steward verifies the claim against the task's definition of done before closing. Done is absorbing - only a human can reopen.

•Priority is human-only - priority:* labels change only by explicit human decision, with the reason logged. Observed behavior may trigger a question, never a silent priority write.

•Uniform label taxonomy - project, task, project:<slug>, owner:<actor>, status:*, and priority:p1-p3 - created idempotently by /project-init.

Autonomous Steward

/project-steward

A scheduled sweep over the portfolio: verifies pending-verification claims, dispatches ready work, escalates tasks that sit unverified past the configured age (default 48 hours), and writes a digest. The install offers cron presets from every-2-hours-business-days down to manual-only.

Projection Sync

/project-reconcile

Keeps read-only projections of the registry in sync on other surfaces. Ships with a Google Tasks adapter (v1) and defines the adapter contract for adding more. Because projections never own state, a lost or stale projection is always recoverable from the registry.

When To Use It

This plugin governs cross-actor work - humans and multiple agents collaborating on projects. For a single agent's own development backlog, use /agent-dev:add-backlog from the agent-dev plugin instead.