design
暂无简介
- License 许可
- MIT
- 最近更新
- 2026-10-26
- Stars
- ★ 0
- Forks
- 0
02 资源文件 6
- 参考文档agents/openai.yaml252 B
- 参考文档references/cli-deliverables.md1.7 KB
- 参考文档references/interview-cli.md2.3 KB
- 参考文档references/interview-mcp.md6.7 KB
- 参考文档references/interview-skill.md2.2 KB
- 参考文档references/mcp-design-workflow.md3.5 KB
03 SKILL.md 正文
Design
In Codex:
$ARGUMENTSis what the user wrote after the skill name.${CLAUDE_PLUGIN_ROOT}and${CLAUDE_SKILL_DIR}mean the plugin folder and this skill's folder (the plugin folder is two levels above this file);agentic-makeris on PATH or is<plugin folder>/bin/agentic-maker. A line that starts with an exclamation mark and a backtick is a command: run it yourself first.
Idea: $ARGUMENTS
Reference files are under ${CLAUDE_PLUGIN_ROOT}/skills/agent-interface-design/references/; the interviews and workflows are in ${CLAUDE_SKILL_DIR}/references/. This runs in the main conversation, so ask the user directly.
Existing surfaces: !${CLAUDE_PLUGIN_ROOT}/bin/agentic-maker detect "${CLAUDE_PROJECT_DIR}" --human 2>&1 | head -20 || true
Procedure
- Choose the surface first. surface-choice.md: a CLI when a mature CLI exists or the agent works in a shell; an MCP server only for per-user auth or tenancy, approvals, typed remote discovery, a shared multi-client service or server-held state; a skill when the gap is workflow guidance; a hybrid when control and execution differ. Label each recommendation as documented fact, benchmark, or practitioner report. Fix an existing CLI before proposing an MCP server.
- Interview. Use the interview for the surface chosen in step 1, in its words, one group of up to four questions at a time: an MCP server, interview-mcp.md (twelve questions); a CLI or script, interview-cli.md; a skill, interview-skill.md. The closed questions go through
AskUserQuestionwith the options the file lists; the open ones are asked as plain text, because that tool offers choices, not free text. Do not put an MCP question to someone designing a CLI or a skill. Stop and propose defaults when the user says "whatever you think is best", and get an explicit yes. - Sketch. Start from the agent's task, then the smallest surface that serves it, then answer the eight contract questions (contract.md). For a CLI: the deliverables in cli-deliverables.md (command grammar, flags, stdout/stderr rules, envelope, exit codes, migration notes). For an MCP server: mcp-design-workflow.md (architecture sketch, tool list with Verb + Resource names, schemas, error strategy, budget).
- Approval gate. Present the sketch and wait. Write no code until the user approves it; then build it with
/agentic-maker:build-cli,/agentic-maker:build-mcpor/agentic-maker:build-skill(each probes what it builds and hands over the derail test), and/agentic-maker:auditand/agentic-maker:derailcheck it afterwards.
If a companion skill for the implementation is installed (the SDK builders are listed in companion-routing.md), name it; do not assume it is.
04 同仓库更多 Skills
derailment-method
暂无简介
targets
暂无简介
derail
Test a skill, plugin, CLI or MCP tool with fresh agents and fix the text that misled them. Started by you; spends usage.
build-skill
Create, redesign or merge a skill, grounded in facts, then hand over the test. Started by you.
build-mcp
Build, extend or repair an MCP server for agents, probe it, hand over the test. Started by you.
build-cli
Build or repair an agent-facing CLI, probe it, hand over the test. Started by you.