v0.7.1 — Improved agent guidance for graph commands

Patch release. Documentation and tool-config changes only — no source, schema, or CLI behaviour changes.

Improved agent guidance for graph commands

The ## Code Graph guidance in the shipped tool configs now describes when to use each command, rather than preferring graph commands over text search in all cases.

• mex graph query and mex graph get lead when a symbol name is known. They are exact and typically 200-500 output tokens. An approximate name can return a confident wrong match, so they should be given exact names.
• mex graph scope is positioned as a starting point for unfamiliar tasks, with an explicit note that it matches on words rather than meaning — a task phrased in vocabulary the code does not use will return weak results, and the manifest should never be treated as a complete answer.
• Agents are told to fall back to Grep/Glob when a scope manifest does not contain what they need, and to rephrase a scope task at most once. The previous wording discouraged text search, which could lead an agent to spend additional calls expanding a manifest that was not going to answer the question.

Applied to AGENTS.md, CLAUDE.md, .cursorrules, .windsurfrules, and .github/copilot-instructions.md.

Note for existing scaffolds

Upgrading does not modify an existing .mex/ scaffold — tool configs are copied at setup time. To pick up the new guidance, replace the ## Code Graph section of your tool config files with the version in templates/AGENTS.md.

New projects get it automatically:

npx mex-agent setup

Upgrade

npm install -g mex-agent@0.7.1

Full changelog: v0.7.0...v0.7.1