Memi 2.6.3: one read-only command before your agent edits UI

[sarveshsea]

Memi 2.6.3 is live with a much smaller first-run path:

npx -y @memi-design/cli@2.6.3 diagnose . --json --no-write --fail-on none

Run it in a frontend repository before a coding agent edits UI. It reads the current codebase, writes no source files, and returns normalized findings with confidence, provenance, and file-anchored evidence. Run the same command after the correction to verify the result.

The first audit needs no account, API key, Figma file, global install, or daemon.

The current focus is verified adoption:

• successful first audits in external repositories
• repeat audits 7 to 21 days later
• public integrations with reproducible evidence
• false positives and missing checks that can become fixtures

Release-day download spikes do not count as retention.

If Memi catches something useful, share the finding ID, framework, correction, and rerun result in Show and tell. If it misses something important, open a fixture-backed issue.

Start here: https://github.com/sarveshsea/memi