agent-autonomy-primitives
от g9pedro
DevOps & CloudКратко
Build long-running autonomous agent loops using ClawVault primitives (tasks, projects, memory types, templates.
Описание
Что делает этот навык
Five composable primitives (typed memory, task files, project grouping, template schemas, heartbeat loops) for building self-directing agent workflows. Agents read their own task queue, execute work, record lessons, and loop without human prompting. Works with ClawVault as the storage layer.
Using typed memory directories and YAML-frontmatter task files keeps agent state queryable and human-readable without a separate database or sync layer.
Когда его использовать
- Setting up a 30-minute cron loop that picks and runs the highest-priority task
- Tracking agent-discovered work as structured task files with priority and due dates
- Grouping tasks by client project so an agent stays scoped to one workstream
- Recording lessons during execution so the same mistake isn't repeated next run
- Building a shared vault where multiple agents coordinate without direct messaging
Пример сценария
User asks: wire up my agent to work through its task backlog every 30 minutes
- 1
Run
clawvault initand create a project withclawvault project add - 2
Add tasks with owner, priority, and project fields via
clawvault task add - 3
Add heartbeat instructions to HEARTBEAT.md: list open tasks, sort by priority and due date, pick top task
- 4
Agent executes the task, then calls
clawvault task done <slug> --reason - 5
If a lesson is learned during execution, agent calls
clawvault remember lesson
Agent runs on schedule, drains its task queue in priority order, and builds a growing memory of decisions and lessons across cycles
Требования
- clawvault CLI (`npm install -g clawvault`)
Установка
clawhub install g9pedro/agent-autonomy-primitivesВыполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.