Agent迁移

Skill Management (skill_manager)

Use this skill when the user asks to add, create, edit, inspect, migrate, or align a Hone skill.

The runtime contract is now:

1. Skills live in skills/<name>/SKILL.md, data/custom_skills/<name>/SKILL.md, or a closer .hone/skills/<name>/SKILL.md.
2. The model sees a compact listing first.
3. The full skill body is only injected into the current turn when skill_tool(skill_name="...") or a user slash command like /<skill-name> is invoked.
4. load_skill is only a compatibility shim. Do not teach it as the primary workflow.

Frontmatter Contract

Prefer this frontmatter schema:

---
name: Human readable name
description: One-line description
when_to_use: Brief trigger guidance
allowed-tools:
  - skill_tool
user-invocable: true
model: optional model override
effort: optional effort override
context: inline
agent: optional agent hint
paths:
  - src/**/*.rs
hooks: {}
arguments: []
script: optional default script path like scripts/run.sh
shell: optional shell hint
---

Notes:

• allowed-tools replaces legacy tools as the main runtime field.
• context should usually be inline; use fork only when the skill should run in an isolated child runner.
• paths hides the skill from the default listing until the active task touches matching files.
• script declares the default executable entrypoint inside the skill directory. skill_tool(..., execute_script=true) can run it with ${HONE_SKILL_DIR} as cwd.
• Keep the Markdown body task-oriented and ready to inject as prompt text.

How To Inspect Skills

When the user asks what skills exist or which skill fits a task:

1. Call discover_skills(query="...") with the user's task or keyword.
2. Summarize the relevant skills from the returned metadata.
3. If one skill should actually be used for the current task, call skill_tool(skill_name="...") so the full skill prompt is expanded for this turn.

When the user asks to inspect a specific skill in detail:

1. Use discover_skills(query="<skill name>") to confirm the match if needed.
2. Call skill_tool(skill_name="<skill id>").
3. Explain the resolved metadata and the injected prompt body, not a hand-written summary that drifts from the source.

How To Create Or Update Skills

When the user wants to create or edit a skill:

1. Collect the intended skill id, description, trigger conditions, and whether users should be able to invoke it directly with /<skill-name>.
2. Write or update the actual SKILL.md file with the new frontmatter schema.
3. Keep the body concrete: trigger rules, required steps, tool usage expectations, and refusal/verification constraints.
4. If you created or changed a skill that should be runnable immediately, validate it by invoking skill_tool(skill_name="<skill id>") and checking the rendered prompt.

Strict Rules

• Do not teach the deprecated skill_tool(action="add" | "update" | "remove") CRUD workflow.
• Do not rely on load_skill as the main user-facing path.
• If a skill is path-gated, mention that it may stay hidden until matching files are involved.
• If runtime enforcement is not implemented for a field such as hooks or strict tool scoping, say so plainly instead of pretending it is active.