brief-me - calibration examples
Eight scenarios. Each shows the failure mode first, then the correct output. Read this when unsure how hard to compress or whether something is a real decision.
The PR base for this repo is
https://github.com/ColaberryIntern/ColaberryEnterprise_AI_LeadershipAccelerator/pull/.
A. Routine progress, nothing needed
Claude built three parts of a feature. It works. Nothing requires Ali.
🚫 Wrong: inventing a question because two file layouts were both viable. 🚫 Wrong: listing all 17 touched files.
✅ Right:
# 🧭 THIS TAB
**Working on:** Per-tenant campaign sending
**Stage:** BUILD
**Branch:** `workstream/tenant-campaign-domains`
**Last meaningful thing that happened:** Campaign sends now read the tenant's own sender config instead of the shared one.
**Next:** Wire the same tenant lookup into bounce handling.
## 🟢 Bottom Line
Three of the four pieces are working. No surprises so far.
## ✅ What Happened
- Campaign sending now uses each tenant's own sender identity.
- Existing lead tracking left untouched, so nothing regressed.
- Unit tests pass.
## 🎯 What I Need From You
**Nothing right now. Claude can continue.**B. PR waiting for approval
🚫 Wrong: "PR #427 is ready for your review."
✅ Right:
## 🔗 Pull Request
**PR #427 - Add tenant campaign domain routing**
https://github.com/ColaberryIntern/ColaberryEnterprise_AI_LeadershipAccelerator/pull/427
**Status:** Checks green, waiting on your approval.
**Why it matters:** This is the last gate before the change can reach production.
## 🎯 What I Need From You
Approve PR #427 (link above). Nothing else is blocked on you.Bare #427 with no link is a defect in this skill, per Ali's standing rule.
C. Genuine architecture decision
Two strategically different designs. This one really is Ali's call.
## ❓ Decision Needed
### Q1. Should every tenant be forced to set up its own sending domain when the tenant is created?
**A. Required at creation**
Cleanest deliverability, but no tenant can be created until DNS is verified.
**B. Optional, added later**
Tenants start immediately on the shared domain and upgrade when ready.
**C. Other**
Write your own answer.
⭐ **Claude recommends: B** - it unblocks tenant creation today and keeps the domain work as a separate, reversible step.
**Reply:** `Q1: B`Contrast: choosing a helper name, a folder, or a test layout is not a decision. That is Claude's, per CLAUDE.md's Autonomy Model.
D. Tests failing
20 pass, 2 fail.
🚫 Wrong: "Testing completed." 🚫 Wrong: pasting the stack traces.
✅ Right:
## 🟢 Bottom Line
The feature works for a single tenant, but **verification is currently failing** - two tests break when two tenants send at the same time.
## ⚠️ Risk
Two campaigns running simultaneously can pick up the wrong sender identity, so a tenant's email could go out under another tenant's domain. Not shippable as-is.
**Verified:** TypeScript, 20 of 22 tests
**Not verified:** concurrent sendingThe word "completed" must never cover a red suite.
E. Story mode
# 🎬 STORY MODE
### Where we started
The goal was to let every tenant send campaigns from its own email domain instead of one shared Colaberry sender.
### What Claude discovered
The campaign engine already knew which account it was working for. The outbound email layer did not - it had one hardcoded sender for everyone.
### What changed
Claude passed the tenant's identity down into the sending path and left the existing lead tracking alone, so open and click history still works the way it did.
### Where we are now
Single-tenant sending works and the tests for it pass.
### The plot twist
Two tenants sending at the same moment can still collide and pick up the wrong sender. That is a real defect, not a rough edge, and it is the next thing to fix.
### What happens next
Lock the sender lookup per send job, then re-run the concurrency tests.
### 🎯 Your decision
You don't need to do anything right now. Claude can continue.Note the plot twist section exists because there was a real problem. Omit it when there isn't one. Never invent one for narrative shape.
F. Five tabs open
The first four lines have to do all the work. Lead with the concrete thing this tab owns ("Per-tenant campaign sending"), never a generic label ("backend work", "the feature"). Branch name is part of the identity - always include it when tabs could be confused.
G. Literal git pull
The local branch is behind origin/main. No approval is involved.
✅ Right:
## 🎯 What I Need From You
**Claude needs the local branch updated from GitHub.** Your working copy is behind, so any build here would test stale code.
`git pull --rebase origin main`No PR URL. Inventing one here is the failure mode this scenario guards against.
H. Merged but not verified in production
# ✅ DONE (code side)
**Built:** Per-tenant campaign sending.
**Verified:** TypeScript clean, 22 of 22 tests pass, PR checks green.
**Merged:** Yes - https://github.com/ColaberryIntern/ColaberryEnterprise_AI_LeadershipAccelerator/pull/427
**Deployed:** 🟡 Not verified - merging to `main` does not deploy by itself.
**Production verified:** 🔴 No. Nobody has sent a real campaign from a tenant domain yet.
**Anything you need to do:** Nothing, unless you want this deployed tonight.
**Where to test it:** Portal > Campaigns, after a deploy."Merged" and "live" are different words for a reason. Collapsing them into "done" is the single most damaging thing this skill could do.