Capabilities
Anything an agent can find and use: a script, skill, command, agent, workflow, memory or knowledge doc.
akm
Agent Knowledge Manager. One searchable library of skills, scripts, workflows, agents and knowledge, usable from any coding agent that can run shell commands. Build it once, use it from Claude Code, OpenCode, Cursor, Aider or whatever comes next. Built by itlackey.
The loop at a glance
Tell akm where your assets already live, index them, then ask for what a task needs in plain language and load the best match by its ref.
akm bundle add ~/.claude # index what you already have, in place
akm index
akm curate "deploy a Bun app" # a curated shortlist for the task
akm show workflows/deploy # the full asset, by ref
akm feedback workflows/deploy --positiveCore concepts
Bundles
A directory of capabilities: a local folder, a git repo, an npm package or a crawled website. Shareable and installable.
One index
Every bundle folds into one local search index. search returns a lean menu, show delivers the full asset.
Refs
A compact handle search returns and show consumes. Get it from search, pass it straight on, never type one by hand.
Why akm?
Every coding agent wants its own copy of your team's knowledge: an AGENTS.md here, a tool-specific skills folder there, prompts scattered across repos and chat logs. akm indexes those assets where they are, loads only what a task needs, and keeps the library independent of any one assistant.
Per-tool folders
Copied, drifting, loaded whole into every prompt.
One akm library
Indexed in place, searched, loaded by ref when needed.
Beyond retrieval
Capture knowledge
akm remember and akm import write into your working bundle, so every agent sees what one of them learned.
Improve with review
Feedback shapes ranking and turns into diffable proposals. Nothing changes in the library without a review.
Durable workflows
Multi-step procedures with gates, retries, budgets and resume, run with akm workflow run.