Workflow guidance for using agent tools effectively — memory, skills, credentials, integrations, collaboration
Workflow guidance for using agent tools effectively — memory, skills, credentials, integrations, collaboration
npx skills add https://github.com/kittyfromouterspace/agentic --skill agent-tools复制此命令并粘贴到 Claude Code 中以安装该技能
Use this skill when working with the Agentic Elixir library. Consult for agent loop configuration, profiles, tools, callbacks, strategies, and testing patterns.
Respect delegation nature, training mode, and human growth — every agent's responsibility
How to discover and use external tools and integrations efficiently — search first, activate for repeated use, keep context lean
| name | agent-tools |
| description | Workflow guidance for using agent tools effectively — memory, skills, credentials, integrations, collaboration |
| core | true |
| loading | always |
| type | skill |
| version | 2.0.0 |
| model_tier | any |
Your tools are self-describing — call them by name and check their schemas for parameters. This guide covers workflows and patterns that aren't obvious from the schemas alone.
Use memory_note to track intermediate results during a task (fast, session-scoped).
Use memory_recall to retrieve recent context from the current session.
Use memory_query with a search query at the start of complex tasks to pull cross-session context.
Use memory_write to persist important decisions or discoveries for future sessions.
When a task matches an installed skill, use skill_read to load its full instructions.
Before installing a skill, always check with skill_info first to preview what it does.
Use get_credential with the service name to trigger the proper OAuth flow.
Never ask the user to paste tokens manually. Authenticated API calls are proxied through
the frontend which injects credentials on your behalf — the secret value is never sent
to your environment.
integration_list() — see enabled integrationsintegration_search(query: "what you need") — find operationsintegration_skill(integration: slug) — get detailed usage guide for complex integrationsintegration_schema(integration: slug, operation: op) — check argsintegration_execute(integration: slug, operation: op, arguments: {...}) — call with auto-injected authFor complex integrations (LinkedIn, GitHub, Slack), always load the skill guide first with
integration_skill — it contains best practices, example calls, and important constraints.
To connect a new API: setup_integration(name: "myapi", spec_url: "https://...", auth_type: "bearer").
Auth is handled automatically — do NOT call get_credential separately.
Use search_tools to discover external tools, get_tool_schema to check args,
use_tool for one-off calls, activate_tool for repeated use. See tool-discovery skill.
Delegate parallel tasks with spawn_sub_agent. Sub-agents run autonomously and report back.
Use target_workspace for cross-workspace tasks, skill to load specific skill instructions.
Max 5 concurrent. Poll with check_sub_agent or list_sub_agents.
Call report_status at natural checkpoints: when you complete a step, find key results,
hit a blocker, or finish your task. This keeps the user and personal agent informed.