cargo-context

Cargo-context lets an agent inspect and edit a workspace's git-backed context repository, the GTM knowledge base of markdown/MDX files, and its runtime sandbox using the Cargo CLI (cargo-ai). The context captures a workspace's go-to-market knowledge such as company narrative, ICPs, personas, plays, proof, and objections, and is read and written by both humans and agents. The skill solves the problem of browsing, reading, writing, and editing these typed context files and inspecting the derived knowledge graph in a controlled way.

The cargo-ai context domain has two subdomains. The runtime subdomain is a checked-out, executable copy of the context repo: cargo-ai context runtime browse lists entries, read (with optional --start-line and --end-line) reads files, write creates or overwrites a file, edit replaces an exact substring, and execute runs shell commands against the sandbox. Critically, write and edit push commits to the default branch immediately (there is no dry-run mode), while execute does not push, so its file changes are discarded, making it suitable for builds, tests, and inspection. Uploaded content files appear read-only under a .files/ directory outside the committed tree and are never pushed. The graph subdomain builds or loads the knowledge graph derived from every markdown/MDX file via cargo-ai context graph get. Because writes push immediately, the skill requires confirming the target workspace with cargo-ai whoami (checking workspace.uuid and workspace.name) before the first write or edit, and asks the user to disambiguate generic workspace names. New files should begin with a YAML frontmatter block setting title and description; frontmatter is not validated but affects graph indexing, and writes can fail with errors like repositoryNotFound, syncConflict, syncFailed, failedToWrite, or deniedPath.

The skill targets GTM and revenue teams and their agents maintaining a Cargo context repository. It requires @cargo-ai/cli (npm) and a Cargo account, authenticating via browser sign-in with --oauth or an API token. It advises editing context derived from sales-call analysis one change at a time with human review rather than batching, and points to companion skills for related tasks such as workspace file uploads and RAG attachments, plus reference files for conventions, response shapes, troubleshooting, and end-to-end recipes.