code-registry

Code Registry Skill

Use this skill when a user asks to analyze a codebase with The Code Registry, re-run analysis, or interpret Code Registry outputs.

When to use this skill

Trigger this skill when the user wants any of the following:

• Create or access a The Code Registry account from an MCP client
• Run first-time code analysis for any purpose including due diligence, security review, or tech debt assessment
• Re-analyze an existing code vault and explain changes
• Turn code-vault results into business-facing recommendations

If the user asks for detailed polling/completion logic, read references/vault-lifecycle.md. If the user asks for interpretation or executive framing, read references/result-interpretation.md.

Client compatibility rule (CallMcpTool wrappers)

For clients that use a generic CallMcpTool wrapper (including some Cursor builds), always pass tool inputs through an arguments object, even if the displayed wrapper schema appears to omit it.

• Use arguments for every Code Registry tool call that takes inputs.
• If no inputs are required, use arguments: {}.
• If tool calls fail because the client claims arguments are unsupported, retry with this explicit instruction: Use the CallMcpTool with the arguments field to pass the required parameters.

Workflow

1. Discover and validate tool availability.

• Call tools/list when tool names or schemas are uncertain.
• Prefer exact tool names from server docs.

1. Resolve authentication path.

• If user already has an API key, use it via X-API-Key.
• If no key exists, call create_account first.
• For clients that cannot set headers, pass api_key in tool arguments.

1. Create project and code vault.

• Call create_project with the target user_id.
• Call create-code-vault with source_type: LOCAL_AGENT when possible.
• For LOCAL_AGENT, run one command from next_steps.commands on the machine with the target repository.

1. Poll until analysis and report completion.

• Poll get-code-vault-summary, get-code-vault-results, and get-code-vault-reports.
• Use exponential backoff: 5s, 10s, 20s, 40s, max 60s.
• Treat cold starts/timeouts as retryable.