akm CLI
akm indexes skills, commands, agents and notes in one place and hands an agent only what a task needs
- 70 ★ on the repository
- Last commit
- MPL-2.0
Extended access
Why this level
- The CLI indexes local folders, git, npm and websites and can run workflows, scripts and agents through a separate command
- A scheduler task like `akm improve` needs activation of entries in the system schedule
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/akmDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The CLI indexes local folders, git, npm and websites and can run workflows, scripts and agents through a separate command
- A scheduler task like `akm improve` needs activation of entries in the system schedule
This is third-party code. Review the repository files before installing.
What it does
akm (Agent Knowledge Manager) is a CLI that gathers into one library what a team has already written for agents: `AGENTS.md` and `CLAUDE.md`, `commands/`, `agents/` and `skills/` folders, standalone Agent Skills packages, workflow files and Markdown wikis. A source can be a local folder, git, an npm package or a website that akm crawls into a local snapshot. Claude Code and OpenCode formats are indexed in place and read-only, while native akm bundles are writable. The loop is: `akm index` builds the index, `akm curate "deploy a Bun app"` returns a shortlist and `akm show <ref>` loads a chosen asset by ref. Search and show only return content; running a workflow, task, script or agent is a separate command. `akm hints --detail brief >> AGENTS.md` tells the agent akm exists. Assets are packaged into bundles with skills, scripts, workflows, agents, instructions and memories, shared through `akm bundle add`. Feedback re-ranks assets right away, and `akm improve` turns edits into reviewable proposals: some are accepted automatically by a quality judge, the rest wait for your decision. Workflows run with steps, gates, retries, budgets and resume; that part is marked experimental.
Who it is for. Teams whose skills, prompts and instructions are scattered across repositories and agents and who want one library for Claude Code, OpenCode and Cursor.
Good fit when
- Every repository has its own `AGENTS.md` and skills folder and you need to search them from one place
- The agent should not carry a huge startup prompt: it takes a shortlist through `akm curate` and loads an asset by ref
- A bundle of skills, scripts and workflows has to reach a colleague through `akm bundle add`
- Negative feedback on an asset should turn into a reviewable fix through `akm improve`
- A repeatable procedure is easier to run as a resumable workflow than to explain again each session
Not a fit when
- You have one agent and a dozen files: plain `AGENTS.md` and skills are enough
- The agent cannot run shell commands: akm works through them
- You need local semantic search on Windows or Linux ARM: the built-in model exists for Linux x64 and Apple Silicon macOS
- You need workflows in a stable form: that part is experimental
Example request
Connect akm to this project, build the index and pick deploy-a-Bun-app materials through akm curateLimitations
- An agent that can run shell commands is required. The npm package needs Node.js 22 or newer, and standalone binaries need no runtime.
- Local semantic search is offered on Linux x64 (glibc) and Apple-silicon macOS: a runtime download of about 86 MB and a model of about 127 MB. Other platforms use keyword search or an external endpoint.
- With no agent connected, work runs on akm's own runtime: the first run downloads a pinned OpenCode 2 binary of about 114 MB.
- Workflows are marked Experimental, and Claude Code and OpenCode formats are indexed read-only.
How to disable. Delete the akm line from `AGENTS.md` by hand and detach sources with the `akm` source and bundle commands. The CLI itself comes off with `npm uninstall -g akm-cli`, and the binary build needs only the file deleted. The index and state live on disk and there is no telemetry.
FAQ
How do I install akm?
Run `npm install -g akm-cli` (Node.js 22 or newer) or download a binary build with `install.sh` from the releases. `akm upgrade` updates in place.
Which agents work with akm?
Claude Code, OpenCode, Cursor, Aider, Windsurf and any assistant that runs shell commands. It is not tied to one agent.
Does akm change my skill files?
Claude Code, OpenCode and standalone Agent Skills packages are indexed read-only. Edits go to akm's own bundles as proposals that can be reverted.
Does akm send data out?
There is no telemetry. The network is used only for sources you configure: Git, npm, websites, registries, your own models and Hugging Face when downloading a local embedding model.
Where do I start after install?
Run `akm setup --yes`, then `akm bundle add github:itlackey/akm-stash` for the official onboarding bundle, `akm index` and `akm curate "your query"`.
Discussion
No questions or reviews yet.
Related
CLI toolsAKB is a PostgreSQL-native knowledge base for agents: documents, tables and files in vaults, search and relations over MCP.
The awiki CLI keeps agent memory in a markdown vault: search before the web and save whatever is worth keeping.
A base AGENTS.md and 12 skills: delegation to Codex, Grok, Devin and Antigravity, planning, copywriting, scoped changes.
The ak command installs, repairs and checks ruflo and agentic-qe, adds the offline RuvNet Brain base and links Claude Code, Codex
One config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
README in short
The README explains why agents need one knowledge library, gives five reasons, then install and a "first useful result" in five commands. A table of supported formats, next steps, a local-first and privacy section and the list of ecosystem repos follow.