# @llodev/pm-tasks-core > Core skill + CRUD vocabulary + autonomous-mode contract shared by every `@llodev/pm-tasks-` adapter. [![npm](https://img.shields.io/npm/v/@llodev/pm-tasks-core?label=npm&color=cb3837&logo=npm)](https://www.npmjs.com/package/@llodev/pm-tasks-core) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE) [![Node.js >= 20](https://img.shields.io/badge/node-%3E%3D20-43853d?logo=node.js&logoColor=white)](https://nodejs.org) [![Agent Skills spec](https://img.shields.io/badge/Agent_Skills-spec-7c5cff)](https://agentskills.io) This package is the shared foundation. **Install it together with at least one adapter** — alone it has no tool-specific formatting and won't activate. ## Install ```bash npm i @llodev/pm-tasks-core @llodev/pm-tasks-trello # or npx skills add llodev/skills/skills/pm-tasks-core llodev/skills/skills/pm-tasks-trello ``` The Claude Code marketplace bundle installs the cascade automatically — see the [repo root README](https://github.com/llodev/skills#install). ## What's inside - **Phases 1–3** of the extraction pipeline (plan input → structured sections → canonical generic card) — the same shape every adapter consumes. - **7 CRUD verbs** every adapter implements: `task.create`, `task.move`, `checklist.check`, `task.close`, `task.due-date.set`, `task.assignee.add`, `task.comment.add`. Plus **3 adapter-optional** verbs (`task.parent.set`, `task.estimate.set`, `task.sprint.set`) — **10 canonical verbs** total. - **Autonomous-mode contract**: activation sentinels (`[autonomous]` / `--auto`), allowlist gate, scope guardrails, audit log format, continuous-loop expectations across multi-task runs. - **Shared init UX library** (`@llodev/pm-tasks-core/init-lib`) consumed by every adapter's `init` script — i18n strings, `promptLocale`, `loadStrings`, platform-aware config dir resolver. - **References** in [`references/`](./references/): `contract.md`, `crud-vocabulary.md`, `autonomous-mode.md`, `generic-card.md`, `audit-log-format.md`, `init-ux.md`. ## Optional — rotate the autonomous-mode audit log When you enable `autonomous` for an adapter, every write-through call appends a JSONL line to `~/.local/share/llodev/pm-tasks//audit.log`. The bundled `rotate-audit.mjs` keeps the log small and emits a JSON status object on each run. ```cron # Daily at 04:00, rotate audit log for Trello + Asana (10 MB limit, keep 12 archives) 0 4 * * * node /path/to/pm-tasks-core/scripts/rotate-audit.mjs --tool trello 0 4 * * * node /path/to/pm-tasks-core/scripts/rotate-audit.mjs --tool asana ``` > [!TIP] > The audit log is the agent's source of truth for "what happened in this autonomous session". Your PM tool (Trello/Asana board) is the **human's** audit log — keep both in sync. See [`references/autonomous-mode.md`](./references/autonomous-mode.md) § _Continuous operation across multi-task loops_. ## License MIT — see [LICENSE](./LICENSE).