corpocode-setup

CorpoCode Setup

With the default native backends, there is nothing to provision — no Python toolchain, no daemon. Installing the plugin already registered CorpoCode's hooks, so it is active immediately. This skill is mostly a verification pass.

Verify it's working (no commands needed)

CorpoCode runs as hooks, so you confirm it by seeing it work, not by running a binary:

1. At the start of a turn you should see an injected <middle-management recommendation> block (moment type, graph-scored files). That is the UserPromptSubmit hook firing.
2. Every hook also appends a line to a project-local log: .corpocode/logs/corpocode.ndjson in the directory you launched the host from. cat the last few lines to see router, verifier, etc.

If you see those, CorpoCode is set up. The one input it needs is an API key for the cheap model, which it reads from your environment — set ANTHROPIC_API_KEY (or OPENAI_API_KEY / GEMINI_API_KEY / OPENROUTER_API_KEY) and you're done. (Or use the anthropic-cli provider, which uses your existing claude login and needs no key, or ollama for fully local.)

Configure config + key without npm (self-contained)

The plugin ships the corpocode binary, so you can run any command without a global install. From the directory you launched the host in:

node "${CLAUDE_PLUGIN_ROOT}/bin/corpocode.js" init      # scaffold ~/.corpocode/config.json + a secrets template
node "${CLAUDE_PLUGIN_ROOT}/bin/corpocode.js" doctor     # health-check

init writes a default config and a ~/.corpocode/secrets file with a placeholder key. Open that file and replace REPLACE_WITH_YOUR_ANTHROPIC_API_KEY with your real key (or set the env var instead — an env var takes precedence over the file). init never overwrites an existing secrets file unless you pass --force, so a real key is always safe.

If ${CLAUDE_PLUGIN_ROOT} isn't set in your shell, the bundled binary lives under Claude Code's plugin cache at ~/.claude/plugins/cache/<marketplace>/corpocode/bin/corpocode.js.

Health check / provisioning — only for some setups

The corpocode CLI is only on your PATH if you installed the npm package (npm install -g corpocode). A plugin-only install runs the bundled binary from the plugin directory and does not put corpocode on PATH — so do not run a bare corpocode … command after a plugin install; it will be "command not found".

• Installed the npm CLI? Run corpocode doctor for a full health check. Under native backends it reports the in-process graph and store; there are no Python/daemon checks to satisfy.
• Deliberately selected the Python backends (backends.knowledgeGraph: "graphify" or backends.contextStore: "openviking" in ~/.corpocode/config.json)? Those are not provisioned automatically. Run corpocode provision (npm CLI required) to install graphify, build the initial graph, generate OpenViking's config, and start the daemon. Then corpocode doctor to verify.

Until a Python graph is built (if you chose one), graph-backed file scoring falls back to a string-overlap heuristic — fail-open by design. The native graph needs none of this: it builds in-process on first use.