Every prompt behind keymail, published whole, secrets redacted. If an AI vendor can see what we asked for, you should be able to see it too.

The policy — what's published, what's redacted, what can be withheld

🤖 keymail is AI-written, and this is the provenance record. Each session below lists every prompt the human typed, complete and in order — the actual working instructions this codebase was built from. Credentials are the only thing removed: secrets, tokens and claim links are redacted in place, and every redaction is annotated with why (legend below).

🤖 What you see today is the complete prompt record. Whole session transcripts — the model's replies, the commands it ran, the tool output — are the same record at ~100× the size, and publishing them is on the roadmap rather than quietly skipped.

🤖 Sessions are in the ledger by default. A whole session can be withheld (it stays on the record: date, prompt count, and a reason, with the content withheld) where redaction can't fix it — for keymail that means anything that could compromise the security of the service itself, because a credential can be redacted but a narrative that maps the defences cannot.

What gets redacted, and why

🤖 Prefer the story to the log? Milestones — the key moments of building keymail, each linked to the session where it happened.

Sessions

Local transcripts prune after roughly a week, so the ledger accrues from each run of tools/prompt-registry.mjs forward. Newest first. Search matches summaries and every prompt inside each session.

WhenModelSummary