一键导入
skill-authoring
Use this skill when an agent wants to codify a recurring delegation pattern as a named, reusable skill and assign it to their own role.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use this skill when an agent wants to codify a recurring delegation pattern as a named, reusable skill and assign it to their own role.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use this skill when reviewing, provisioning, or documenting MCP membrane tools for external clients such as Perplexity, Claude, or Codex. It keeps Muninn capture, LifeGraph tools, and raw graph/runtime tools separated.
Use this skill when authoring or revising an agent role lens before governed execution. It gathers the missing role definition inputs, explains the required payload shape, and prepares the payload for the role.create_or_update workflow.
Use at session start and closeout to retrieve and write durable Muninn continuity context.
Use when the user needs planning, decomposition, orchestration, or seam selection guidance.
Use when the user needs validation strategy, test selection, or confidence reporting.
Use this skill when an agent summarizes the operator's LifeGraph, answers "what is in my LifeGraph?", or turns LifeGraph records into planning context. It requires provenance-aware summaries that separate confirmed graph facts from seeded placeholders, inferred intent, and recommended next structure.
| name | skill-authoring |
| description | Use this skill when an agent wants to codify a recurring delegation pattern as a named, reusable skill and assign it to their own role. |
| catalog | {"skill_name":"skill.authoring","implied_tools":["skill.list","skill.register","skill.assign"],"validation_state":"validated","skill_markers":["governed","self_directed"],"field_sources":{"required_fields":["skill_name","description","subagent_kind","goal"],"optional_fields":["allowed_tools","allowed_classes"],"repo_skill_path":"skills/skill-authoring/SKILL.md","workflow":"skill.list → skill.register → skill.assign"}} |
Use this skill when you have identified a recurring delegation pattern in your work and want to register it as a named, reusable skill that you can assign to your roles.
Do not create a skill for a one-off task. Skills are for patterns.
skill.list to check if a similar skill already existsskill.register with the full payloadskill.assign with the role_name of the role that should have access to the skillskill.register fields| Field | Required | Notes |
|---|---|---|
skill_name | Yes | Lowercase, hyphens/underscores only, max 64 chars. Stable identifier. |
description | Yes | What the skill does and when to use it. 1-3 sentences. |
subagent_kind | Yes | The worker role that executes this skill (almost always philote-worker). |
goal | Yes | The goal template injected into the subagent at invocation. May include {{placeholder}} variables. |
allowed_tools | No | Explicit tool IDs the subagent may use. If omitted, falls back to class allowances. |
allowed_classes | No | Tool class names allowed (e.g. workspace, utility, shell). |
{
"skill_name": "file-summarizer",
"description": "Summarize a workspace file into bullet points for quick review.",
"subagent_kind": "philote-worker",
"goal": "Read the file at {{file_path}} and return a structured bullet-point summary of its key content."
}
{
"skill_name": "deep-search",
"description": "Delegate a multi-source research task to a focused subagent. Use when a topic requires reading multiple workspace files or documents before synthesizing an answer.",
"subagent_kind": "philote-worker",
"goal": "Research {{topic}} thoroughly. Read all relevant files in the workspace. Return a structured summary with: key findings, open questions, and recommended next steps.",
"allowed_tools": ["workspace.read", "workspace.list"],
"allowed_classes": ["workspace", "utility"]
}
After skill.register succeeds, assign the skill to your current role:
{
"role_name": "orchestrator",
"skill_name": "deep-search"
}
skill.assign is scoped to your own agent — you cannot assign skills to other agents' roles.
bash.exec in your roleallowed_tools lists — subagents should not have more authority than the task requirescode-reviewer not reads-files-and-gives-feedback