Help for the chat workflow

The primary surface for maintainability-agent is an AI chat host driving the local MCP process. After configuration is answered and the user chooses to run, the deterministic audit measures the repository and returns a bounded work order; the host shows it to the coding agent without granting permission for unrelated refactoring. The CLI is the automation and CI surface. Chat, MCP, and an interactive CLI TTY are one setup: the same questions.

Start with the part of the conversation you need:

• First run and questions — setup, repository grants, history consent, and when a report file may be saved.
• What the analyzer pool runs — analyzer-primary evidence, built-in fallback, selection, and missing-tool remedies.
• Reading the report and its history — estimate, range, grade, work order, recurrence, baselines, economic context, and why declaration rates are withheld for an unparsed language.
• Language support — the sixteen parsed languages, how each decides where a declaration ends, and what each one deliberately misses.

For integration details, see IDE and agent integration. For the binding product promises, see product intent.