Can Codebuff users falsify a no-write repo-context proof?

[nike-17]

Disclosure: I maintain Sverklo.

Codebuff's multi-agent file selection and planning is a useful comparison point for the claim I am trying to test: before an agent edits, can a local no-write proof show the repo relationships that matter?

I am not asking for stars or adoption. I am looking for one correction from people who already use coding agents.

From a repo root:

npm exec --yes --package=sverklo@latest -- sverklo prove --no-write --guided --markdown

Useful feedback would be one of:

• the proof picked the wrong central files
• the selected symbol was noisy or unimportant
• callers were technically references but not meaningful impact
• Codebuff already finds the right files better
• grep/ripgrep was plainly the better tool
• setup friction around Node/model cache/output

Public proof thread for receipts/corrections: sverklo/sverklo#79

Tracker for the proof sprint: sverklo/sverklo#85