GitHub
Like Skills, but executable. Packages turn procedural know-how — how to deploy, provision, validate, and tear down — into software your agent, your shell, and your CI can all run.
A package is a named, reusable unit of operational know-how — how to provision a server, configure DNS, render a Terraform plan, validate a deployment, and safely tear it down again.
Skills give that knowledge to an agent. Packages give it to everyone: agents, humans in a terminal, CI pipelines, and scripts. Same procedure, one executable contract.
$ npx bc-pkg bigconfig-ai/once@typescript package validate
$ npx bc-pkg bigconfig-ai/once@typescript package create
$ npx bc-pkg bigconfig-ai/once@typescript package deleteOne package. One contract. Every runtime.
Both start from the idea that procedural knowledge should be reusable. They diverge on where that knowledge lives and how it runs.
| Question | Skill | BigConfig package |
|---|---|---|
| What is it? | Agent-native procedural knowledge | Shell-native procedural knowledge |
| Who can run it? | An agent with skill support | Humans, agents, CI, scripts, containers |
| Does it support library reuse? | Not as a first-class model | Yes, through normal package dependencies |
| What is the interface? | Agent instructions | Executable commands and workflows |
| What does it optimize for? | Making an agent better | Making a procedure runnable and reusable |
Read the long form: Skills vs. BigConfig Packages →
Skills make agents smarter. BigConfig makes procedures runnable. The same package runs from your terminal, your agent, your CI job, your cron, or another script — not just inside one chat product.
Packages are real software packages. They depend on libraries, share validation logic, abstract cloud providers, and version like code. Procedural knowledge stops drifting and starts compounding.
Every package exposes the same verbs — validate, build, create, describe, delete — so credentials and authority can be scoped per action. Agents can safely describe and validate while humans stay in the loop for create and delete.
Packages render real files: Terraform plans, Ansible inventories, app configuration, lock files, validation reports. The outputs outlive the conversation and can be reviewed, diffed, and committed.
A package is not trapped inside an agent. It is a normal command-line program. Anyone — or anything — that can run a shell command can run it.
The agent inspects, describes, validates, and recommends — and asks before destructive verbs.
Operators run the same commands during incidents and routine work — no chat client required.
Pipelines validate on every PR and run create or delete on protected branches with scoped credentials.
Same binary, scheduled or wrapped. The package is just a command — anything that calls a command can call it.
Every BigConfig package exposes the same five verbs. The credentials each one needs are part of the contract — so your agent can safely validate and describe with read-only access while you keep create and delete behind human approval.
- run npx bc-pkg bigconfig-ai/once@typescript package validate
- iterate on failure