Version. 0.1 · Date. 2026-08-06 · Author. Claude.ai (drafting) / Marvin Percival (approving).
Target. /Users/dunin7/loomworks-engine, branch main.
Baseline. Post CR-2026-177. CC confirms exact HEAD at Step 0.
Companion to. standing-notes/dunin7-build-list-v0_29 (B-30); whatever record documents B-27's original diagnosis — CC locates these at Step 1.
Charter. Autonomous regime. Status. Ready for CC execution, with a scoped Step 1.
The build list records B-30 as: "Never repaired; only the broken door was removed. The test suite cannot see it." B-27 was closed by removing the symptom, not the underlying trap — and the suite has no way to notice if the same trap catches something else.
This CR's §2 is deliberately thin. Claude.ai does not have the detail of what the trap actually is; that lives in B-27's original diagnosis in the record. Step 1 reads it. This is a scoped inspection inside the CR rather than a guess written into it.
Step 1 — read the actual defect (read-only). Locate B-27's diagnosis in the record and the commit that closed it. Determine precisely: what the trap is, why removing the door didn't repair it, and what a second occurrence would look like. Report this back before proceeding.
Step 2 — repair. Fix the underlying trap, per what Step 1 found.
Step 3 — make it visible to the suite. A test that fails if the trap is reintroduced. This is the part the build list flags as missing — a repair without it leaves the same blind spot.
Halt condition. If Step 1 finds the repair is larger than one or two files, or needs a design decision (e.g. the trap is structural rather than a bug), halt and report — it leaves this CR and gets its own scoping note, same treatment B-54 got.
| Step | What | Mode |
|---|---|---|
| 0 | Pre-flight. | Auto |
| 1 | Read B-27's diagnosis; characterize the trap. Halt if not small. | Auto, conditional halt |
| 2 | Repair. | Auto |
| 3 | Regression test that would catch reintroduction. | Auto |
| 4 | Full suite; ruff 0; mypy at/below baseline. | Auto |
| A | Checkpoint. | Checkpoint |
| 5 | Tag cr-2026-178-b30-regression-trap. Push. | Auto |
Execute CR-2026-178 v0.1 at ~/Downloads/cr-2026-178-loomworks-b30-
regression-trap-v0_1.md. Confirm the CR number against the ledger first.
B-30: the trap that caused B-27 was never repaired — only the broken
door was removed — and the test suite can't see it.
Step 1 is a read, not a build: find B-27's original diagnosis in the
record and the commit that closed it, and report what the trap actually
is before touching anything. The CR is deliberately thin on §2 because
that detail isn't in Claude.ai's hands — it's in the record.
HALT after Step 1 and report if the repair turns out to span more than
a file or two, or needs a design decision. It leaves this CR and gets
its own scoping note rather than growing this one.
Step 3's regression test: deliberately break it once to prove it
catches, then revert — same discipline as the CI gates.
Baseline: post CR-2026-177. Confirm current HEAD at Step 0.
DUNIN7 — Done In Seven LLC — Miami, Florida CR-2026-178 — B-30 regression trap — v0.1 — 2026-08-06