KeyBank

Tell an agent “use the development key.” It matches that phrase to a catalog entry and either writes the key into a .env file or sets it for the command it is about to run.

The same skill can add a new API key from a description. The only thing you paste yourself is the secret, into one file: ~/.keybank/secrets.env.

Quick start

1. Install the CLI.

pipx install agent-keybank

uv tool install agent-keybank works too.

2. Run setup once. It creates ~/.keybank and asks which coding agents you use, then installs the skill for those agents.

keybank setup

Non-interactive:

keybank setup --agents all

keybank setup --agents claude,codex,cursor

3. Tell an agent what secrets you need. It writes the ids and descriptions. You paste each secret into ~/.keybank/secrets.env. After that, “use the development key” is enough.

After install

Say something like: “I need a development key and a production key for my API, both exposed as SERVICE_API_KEY.”

The agent adds catalog entries with ids, descriptions, and the runtime name. It does not see the secrets.

Open ~/.keybank/secrets.env and paste the values yourself:

service-prod=...
service-dev=...

Do not ask the agent to open that file. Agents must never read it.

From then on you can say:

• “Use the development key and put it in this folder’s .env.”
• “Run this script with the production key.”
• “Add a staging key that maps to SERVICE_API_KEY.”

Where keys live

The bank is on the machine, not in any git repo. Override the location with KEYBANK_HOME.

catalog.yaml never contains secrets. secrets.env is mode 600 and stores values under the catalog id:

service-prod=...
service-dev=...

What setup does

keybank setup creates ~/.keybank/ with catalog.yaml and secrets.env if they are missing, then copies the skill into the agent folders you chose.

--agents all covers Claude Code, Codex, Cursor, OpenCode, Gemini, Factory, Grok, and Amp.

Re-run keybank setup to refresh the skill. The bank stays put. pipx upgrade agent-keybank (or uv tool upgrade agent-keybank) refreshes the CLI.

pipx uninstall agent-keybank

Uninstall removes the CLI. It does not delete ~/.keybank.

Commands

Discover keys. These print ids and descriptions, never secret values.

keybank list

keybank list --json

keybank show development

keybank resolve "personal project"

Load a key into a folder as the runtime name the process expects.

keybank load development --into ./scripts/my-probe/.env

keybank load service-dev platform-work --into ./scripts/my-probe/.env

keybank load service-dev --as OTHER_API_KEY --into ./scripts/my-probe/.env

--into DIR writes DIR/.env. --as VAR renames the secret for one key.

Run a command with the key in the process environment and skip writing a file.

keybank run development -- python script.py

Add a catalog entry. Then paste the secret into ~/.keybank/secrets.env. You can also run keybank set-secret <id> and type it at a hidden prompt.

keybank add service-dev \
  --description "Development environment" \
  --alias dev --alias development \
  --maps-to SERVICE_API_KEY \
  --public SERVICE_API_URL=https://api.dev.example.com

Other commands: keybank init, keybank remove <id> --yes, keybank doctor, keybank home.

Catalog format

Agents may read this file. Keep secrets out of it.

keys:
  - id: service-prod
    description: Production API, personal project
    aliases: [prod, production]
    maps_to: SERVICE_API_KEY
    public:
      SERVICE_API_URL: https://api.example.com

  - id: service-dev
    description: Development environment
    aliases: [dev, development]
    maps_to: SERVICE_API_KEY
    public:
      SERVICE_API_URL: https://api.dev.example.com

maps_to is the environment variable the process should see. That is how two keys both become SERVICE_API_KEY at use time. public holds non-secret companions such as a base URL.

Prefer letting an agent run keybank add. Hand-edits work if you stick to this shape.

Agent skill

The skill is one file bundled with the CLI. It follows the Agent Skills standard. keybank setup copies it into the personal skill folders for the agents you pick.

New agent sessions pick it up from those folders. The skill tells the agent to use the CLI and to stay out of secret files.

Security rules

The CLI is the only process that should open secrets.env. It never prints secret values.

Generated .env files are mode 600. Add .env to .gitignore in the folder you load into.

keybank run is the safer option when you do not need a leftover file.

Do not put a master .env in a git repo root and do not ask an agent to open it.

Tests

python3 -m unittest discover -s tests -v

License

MIT. See LICENSE. Anyone can use, copy, modify, and distribute this, including commercially.