aikito
Install, configure, and operate Aikito workspaces, including memory, skills, project resources, MCP servers, subagents, adoption, synchronization, and status verification.
lsaint · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add lsaint/aikito --skill aikito技能说明
name: aikito description: Install, configure, and operate Aikito workspaces, including memory, skills, project resources, MCP servers, subagents, adoption, synchronization, and status verification.
Aikito
Purpose
Use Aikito as the canonical, Git-managed source for durable Agent resources and synchronize those resources into supported coding agents and projects.
Mental Model
Keep the CLI source and user workspace separate:
~/aikito-src CLI source checkout
<workspace> canonical user workspace (defaults to ~/aikito)
The workspace contains the sources of truth:
<workspace>/
├── agents.toml
├── skills.toml
├── subagents.toml
├── mcps/
├── global/AGENTS.md
├── skills/
├── memory/
└── projects/<name>/
Agent configuration directories and project .agents/ directories are runtime
entry points, not independent sources. Modify canonical workspace files first,
then synchronize them with the CLI.
Standard lifecycle for new canonical resources:
add ↓
edit ↓
show ↓
sync ↓
status / diff / doctor
Existing external resources follow a separate import path:
adopt ↓
canonical resource ↓
show / edit / sync / status
Resolve <workspace> with aikito path workspace before accessing canonical
files directly. AIKITO_DIR can temporarily override the persisted workspace.
Use the installed aikito command rather than assuming the CLI lives inside
the workspace.
Bootstrap
When the CLI is not installed, use the official source repository and keep its checkout separate from the future workspace:
git clone https://github.com/lsaint/aikito.git "$HOME/aikito-src"
export PATH="$HOME/aikito-src/bin:$PATH"
Before cloning, verify that the target does not contain unrelated user data. Do not overwrite an existing checkout without the user's direction.
Initialize and inspect a workspace:
aikito init workspace ~/aikito
aikito status
aikito init workspace refuses the CLI source tree, another source checkout,
and an unrecognized non-empty directory. Do not try to bypass these guards.
Initialization detects locally installed Agents and writes only those entries to
the root agents.toml, which is the Agent source of truth for all projects.
After initialization, inspect the generated files before synchronizing
anything.
Safety Protocol
- Start with read-only discovery: inspect relevant canonical files, run
aikito status, and use resource-specific status commands. - Run
aikito adoptwithout--applyfirst. Show the user the adoption plan, conflicts, and credential handling before applying it. - Use
--dry-runfor synchronization commands that support it. - Treat unmanaged targets and conflicting instructions as decisions for the user. Do not silently overwrite them.
- Keep API keys, tokens, passwords, and OAuth material out of the Aikito workspace and Git. Store environment-variable references in canonical MCP configuration.
- Warn before any explicit force or prune operation and scope it to the reviewed target.
- Do not treat a local Git repository as safe to publish. Review memory and configuration for private data before adding a remote or pushing.
- Preserve unrelated user changes in both the workspace and target Agent configuration.
Resource Routing
Global Instructions and Skills
Canonical sources:
<workspace>/global/AGENTS.md
<workspace>/skills.toml
<workspace>/skills/<skill-name>/
Create, review, and synchronize skills:
aikito add skill <name> [--description <desc>]
aikito edit skill <target>
aikito show skills
aikito show skill <target>
aikito sync global --dry-run
aikito sync global
aikito status
Do not edit generated Agent-native instruction or skill entries as independent sources.
When adding, renaming, or removing a global skill, treat the change as a cross-runtime migration rather than a directory-only edit:
- Update
skills.tomland references to the old skill name. If the skill is a bundled CLI resource, update its source and init template in the CLI source checkout separately; the user workspace is never a release source. - Verify that every supported Agent that loads global skills has the correct
skills_pathinagents.tomland in theaikito inittemplate. Do not assume that an Agent CLI and its IDE use the same global skill directory. - Run the relevant tests, then run
aikito sync globalandaikito status. - Inspect each managed Agent target and confirm that the new skill is visible and any stale link for the old name is gone.
- Keep a global skill out of project
agent.tomlfiles unless the project intentionally overrides or copies it.
Projects
A workspace is the central source of truth and normally exists once. A project registration represents one code directory and its project-specific Agent resources. When a requested instruction, skill selection, or memory is project-specific and the current code directory is not registered, ask whether the user wants to register it. Do not register a project for a global resource.
After confirmation, initialize the project from its code directory:
aikito init project
The directory name and current path are the defaults. Use explicit arguments when needed:
aikito init project <name> <path>
This command creates the canonical project skeleton and synchronizes its
.agents/ runtime. Treat existing unmanaged runtime resources or a project
name bound to another path as conflicts for the user; do not bypass them.
Project configuration belongs under:
<workspace>/projects/<name>/agent.toml
<workspace>/projects/<name>/AGENTS.md
<workspace>/projects/<name>/memory/
Synchronize and verify with:
aikito show projects
aikito show project <name>
aikito sync project <name> --dry-run
aikito sync project <name>
aikito status
aikito diff
aikito show memory
Project instructions and memory remain linked to canonical sources.
阅读完整来源 ↗