From Clear Ideas
Use when the user wants Claude Code to create, edit, validate, clone, schedule, test, or run Clear Ideas agents through the Clear Ideas MCP tools.
How this skill is triggered — by the user, by Claude, or both
Slash command
/clearideas-claude:create-edit-agentsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when the user wants Claude Code to do agent authoring work for Clear Ideas. Treat Claude Code as the flexible agent editor and Clear Ideas as the schema, validation, persistence, and runtime authority.
Use this skill when the user wants Claude Code to do agent authoring work for Clear Ideas. Treat Claude Code as the flexible agent editor and Clear Ideas as the schema, validation, persistence, and runtime authority.
Clear Ideas MCP tools use the word agent for authoring and execution. Use agent in user-facing text unless the user explicitly uses a different term. The saved agent JSON still has runtime schema fields such as prompts, variables, triggers, and schedules.
Do the drafting and repair work in Claude Code:
Do not hand-author agent JSON from memory when the MCP schema or a saved agent can be read. Use get_agent_definition_schema when step fields, model aliases, skill IDs, schedule shapes, examples, deprecations, or current limits matter.
list_agent_connections.start_agent_design to get a strong initial draft and durable design session.continue_agent_design.validate_agent_definition with the draft.validate_agent_definition until there are no blocking issues.create_agent with the final draft.If the user explicitly asks Claude Code to build from scratch, Claude Code may draft the agent JSON directly, but still must use get_agent_definition_schema when uncertain and validate_agent_definition before saving.
list_agents and ask the user to choose.get_agent for the selected agent. Keep its authoringHash.validate_agent_definition with the edited draft.update_agent, passing the complete edited agent and expectedHash from get_agent.get_agent again and summarize what changed.Use direct update_agent patches for obvious metadata-only changes such as name, tags, status, simple schedules, or a single clearly specified field. Use the full draft flow for steps, variables, outputs, triggers, HITL, webhooks, sub-agent invocations, source connections, or dataflow.
Validation is the authority. When validate_agent_definition returns blocking issues:
Never save a structural edit with validation disabled unless the user explicitly asks to bypass validation and understands the risk.
When the user wants to test or debug execution:
run_agent with runNow: true and the required runtime variables.get_agent_run and get_agent_run_events to inspect status, outputs, and runner events.control_agent_run with action: "respond" after the user supplies answers or a decision.control_agent_run with action: "cancel" or action: "run".For schedules and variable presets:
get_agentschedule_agent for schedule/variable preset updatesprompts as the canonical top-level step array.{{variableName}} references for agent variable and step-output dataflow.outputContract, skillIds, autoAddSkills, trigger mappings, schedules, and source connections unless the requested edit affects them.get_agent_definition_schema for current built-in IDs.requiresOverride: true.get_agent_definition_schema and validate_agent_definition.When work is complete, include:
npx claudepluginhub clearideas/claude --plugin clearideas-claudeGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.