From 3f78c61cbea394752d54dc71ea3892e464047800 Mon Sep 17 00:00:00 2001
From: Drew Ritter <drew@primeradiant.com>
Date: Thu, 6 Aug 2026 18:08:44 -0700
Subject: [PATCH] fix(sdd): land proven e2e scenario cards

Require spec-derived E2E artifacts to be authored, independently checked, and run before an exact-path commit. Add conditional bootstrap approval, focused artifact review, a separate correction/rerun loop, and clean HEAD/path durability checks before finishing. Preserve existing product-code failure handling.
---
 .../spec-derived-e2e.md                       | 44 ++++++++++++++++---
 1 file changed, 39 insertions(+), 5 deletions(-)

diff --git a/skills/subagent-driven-development/spec-derived-e2e.md b/skills/subagent-driven-development/spec-derived-e2e.md
index 10359cfd..b7146070 100644
--- a/skills/subagent-driven-development/spec-derived-e2e.md
+++ b/skills/subagent-driven-development/spec-derived-e2e.md
@@ -24,11 +24,45 @@ plan touches.
 
 Use superpowers:agentic-end-to-end-testing:
 
-1. Dispatch a card-author subagent per its authoring-cards-from-a-spec.md.
-2. Run its scripts/check-cards-against-spec yourself on the author's output
-   — self-attestation is not the gate.
-3. Dispatch a runner subagent per its runner-prompt.md against the built
-   branch.
+1. Require a clean working tree (`git status --porcelain` is empty) before
+   dispatching the card author. Record `E2E_BASE=$(git rev-parse HEAD)` before
+   any card or bootstrap-spec work.
+2. Dispatch a card-author subagent per its authoring-cards-from-a-spec.md. Ask
+   the author to report every authored card path and, on the bootstrap path,
+   the governing spec path it changed.
+3. Run `scripts/check-cards-against-spec` yourself on the author's output —
+   self-attestation is not the gate. Preserve the complete checker output and
+   compare the resulting paths with the author's report.
+4. If the author changed the governing spec, present the spec-only diff to
+   your human partner and wait for explicit approval before continuing. This
+   approval gate is conditional: a pre-locked, already-approved scenario table
+   does not require a new approval.
+5. Dispatch a runner subagent per its runner-prompt.md against the built branch
+   and the checked cards while those artifacts remain uncommitted. Retain the
+   runner report, ledger, and evidence; a successful card is not yet a shipped
+   artifact.
+6. After every card passes, stage only the card paths reported by the author
+   plus the approved bootstrap spec path, if any. Do not stage product code or
+   unrelated evidence. Commit the unchanged successful artifacts with:
+
+   ```text
+   test(e2e): add spec-derived scenario cards
+   ```
+
+7. Dispatch a focused artifact reviewer for that commit, using the governing
+   spec, author report, independent checker output, and runner evidence. The
+   reviewer checks the exact artifact paths, their unchanged spec assertions,
+   and the evidence for every passing card.
+8. If the focused review requires an artifact change, dispatch one separate
+   artifact-fix subagent with the review findings. The original card author and
+   focused reviewer remain finders, not fixers. Rerun the checker and every
+   affected card, stage only the corrected artifact paths (and approved
+   bootstrap spec path, if applicable), commit them with the same message, and
+   repeat the focused artifact review over the new commit range.
+9. Before invoking superpowers:finishing-a-development-branch, require empty
+   `git status --porcelain` output and verify every authored path is tracked in
+   `HEAD` with `git ls-files --error-unmatch -- "$path"`. Report `E2E_BASE`,
+   the artifact commit range, reviewer verdict, and exact durable paths.
 
 ## Failure handling