Agent Skills › lthoangg/OpenAgentd › self-healing

self-healing

GitHub

用于动态修改 Agent 自身配置，包括切换模型、安装或编写技能、添加 MCP 服务器及插件等。通过读取参考文档并应用最小化补丁来更新配置文件，实现自我修复和扩展，不涉及用户项目代码。

appv3/contract/builtin_skills/self-healing/SKILL.md lthoangg/OpenAgentd

Trigger Scenarios

要求切换模型或思考级别请求安装新技能或插件要求添加 MCP 服务器要求生成图片/视频配置要求升级 Agent 自身能力

Install

npx skills add lthoangg/OpenAgentd --skill self-healing -g -y

More Options

Non-standard path

npx skills add https://github.com/lthoangg/OpenAgentd/tree/main/appv3/contract/builtin_skills/self-healing -g -y

Use without installing

npx skills use lthoangg/OpenAgentd@self-healing

指定 Agent (Claude Code)

npx skills add lthoangg/OpenAgentd --skill self-healing -a claude-code -g -y

安装 repo 全部 skill

npx skills add lthoangg/OpenAgentd --all -g -y

预览 repo 内 skill

npx skills add lthoangg/OpenAgentd --list

SKILL.md

Frontmatter

{
    "name": "self-healing",
    "description": "Change the agent's own OpenAgentd setup on request — model or thinking level, tools, member (delegate) agents, MCP servers, skills (install from a URL or write one), plugins, and image\/video generation. Use for requests like \"switch your model to X\", \"install this skill\", \"add the GitHub MCP\", \"write a plugin that…\", \"use Gemini for images\", \"upgrade yourself\". Not for the user's own project code."
}

Self-Healing

This file is an index. Pick the topic, then read only that reference — each one holds the file formats, recipes, and when changes take effect.

References

Topic	Read	Covers
Agents	`{SKILL_DIR}/references/agents.md`	`code.md` and member agents: model, thinking level, tools, system prompt
MCP servers	`{SKILL_DIR}/references/mcp.md`	`mcp.json` format, Settings → MCP, when servers start
Skills	`{SKILL_DIR}/references/skills.md`	install from a URL, write a new skill, discovery order, overrides
Plugins	`{SKILL_DIR}/references/plugins.md`	provider and tool plugins, where they live, reload, review
Plugin API types	`{SKILL_DIR}/references/openagentd.d.ts`	full `openagentd` module typings — read when writing a plugin
Image / video	`{SKILL_DIR}/references/media.md`	`multimodal.yaml` for `generate_image` / `generate_video`

Paths

References use these placeholders; substitute the real paths:

Placeholder	Path
`<CONFIG_DIR>`	`{OPENAGENTD_CONFIG_DIR}`
`<AGENTS_DIR>`	`{AGENTS_DIR}`
`<SKILLS_DIR>`	`{SKILLS_DIR}`

Rules for every topic

Change configuration files only — no OpenAgentd code changes.
Never restart OpenAgentd from shell; it ends the current turn. When a change needs a restart, tell the user.
Use absolute paths; file tools refuse ~.
Secrets: .env is off-limits to the file tools. API keys and logins go through Settings → Providers or openagentd auth <provider>. Never write a secret into a config file without the user's explicit consent.
This skill's own files are read-only and regenerated by OpenAgentd. To customise a bundled skill, write one with the same name under {SKILLS_DIR} — it takes precedence.

Workflow

Read the reference for the topic, then read the target file.
Compute the minimal diff — change only what was asked; never reformat unrelated lines.
Show a diff block with a one-line summary.
Wait for confirmation unless the request was already explicit — then apply in the same turn after showing the diff.
Apply with patch (it creates parent directories), then read the file back.
Report what changed, which file, and when it takes effect.

Failure modes

File not found → confirm the name; list the directory to show what exists.
Malformed frontmatter or JSON/YAML → show the problem; don't repair unless asked.
Request outside these topics (code changes, OS setup) → say so and suggest the right path.