Merge pull request #680 from digipulse-engineering/contrib/sync-1787259346

sync: update 3 file(s) in core/

Author: Fr-e-d

.gaai/core/agents/sub-agents/qa.daemon-prompt.md

@@ -127,6 +127,28 @@ form or write down your `plan_conformance` conclusion.
      identifies the current replacement/migration AND continued use materially impairs
      security, reliability, interoperability or supported maintenance — even if not yet
      formally deprecated.
+
+   > **Inventory `blocking` is state-of-the-art-only — never use it to record a scope/plan
+   > problem.** This inventory classification exists solely to answer "does this surface, on
+   > its own technical merits, breach the currentness/materiality floor" — and `blocking` here
+   > is the ONLY thing that obligates a matching `evidence[]` entry (step 2) with a named
+   > primary authority. A `plan_conformance` defect — an undisclosed diff, an out-of-scope
+   > change, code the PLAN said was untouched, anything that contradicts the Story/PLAN/Epic —
+   > is a completely different kind of problem: there is no "official standard" to cite for "the
+   > plan said X and the diff shows Y." Record it as a `findings[]` entry with
+   > `classification: blocking` and `root_cause: plan | implementation` (below) — that alone
+   > fails `plan_conformance`, and it needs **no** `evidence[]` entry. Leave the surface's own
+   > `changed_surface_inventory` classification at whatever its currentness merits actually are
+   > (usually `non_blocking`) unless it *also* independently breaches the materiality floor.
+   > Marking a surface `blocking` in the inventory for a pure scope/plan reason forces an
+   > authority-evidence obligation you cannot satisfy, and the validator will reject the whole
+   > handoff as `QA_HANDOFF_INVALID: blocking finding '<id>' ... has no supporting evidence` —
+   > a real observed failure mode: a genuine scope-violation finding, correctly identified,
+   > repeatedly rejected on handoff because its surface was inventory-classified `blocking`
+   > with no authority to back it. The correct shape: a `findings[]` entry with
+   > `classification: blocking`/`root_cause: plan` naming the surface, that surface's
+   > `changed_surface_inventory` entry left `non_blocking`, and no evidence entry for that
+   > finding at all.
 2. **Gather primary-authority evidence for every non-N/A surface.** Primary authority is
    closed to: an official standard/regulator publication, official vendor/runtime/framework
    documentation or advisory, official maintainer release/security notes, or a product-owned

.gaai/core/skills/delivery/qa-review/SKILL.md

@@ -51,7 +51,13 @@ Activate after implementation is complete. This is a **hard quality gate**.
 - Only files within Story scope were modified
 - No feature creep introduced
 - No unrelated refactors included
-- Unexpected changes → FAIL
+- Unexpected changes → FAIL, recorded as a `findings[]` entry (`classification: blocking`,
+  `root_cause: plan | implementation`) that fails `plan_conformance` on its own. This is a
+  different mechanism from the `changed_surface_inventory` `blocking` classification used in
+  Step 7 — do NOT mark the affected surface `blocking` in the inventory or attach an
+  `evidence[]` entry for a pure scope/plan violation; there is no primary authority to cite for
+  it, and doing so produces an unsatisfiable handoff rejection (see `qa.daemon-prompt.md`
+  Two-Axis Review step 1 for the worked example).
 
 ### 3. Rule Enforcement
 - Confirm compliance with each applicable rule
@@ -99,6 +105,9 @@ TS|Error|warning" /tmp/tsc-output.txt | tail -100`.
 
 Independent of Steps 1-6, evaluate and record `state_of_the_art_conformance` — do this
 BEFORE finalizing `plan_conformance`, to avoid anchoring on "tests passed" (DEC-200 D1).
+This step's `changed_surface_inventory` `blocking` classification (and its `evidence[]`
+obligation) is scoped to currentness/materiality only — a Step 2 scope violation never
+belongs here, see Step 2's note above.
 
 - **Materiality floor.** A functionally correct implementation that passes every business
   test is still `FAIL` on this axis when any changed surface is `blocking`: a named primary

CHANGELOG.md

@@ -10,6 +10,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [Unreleased]
 
 ### Changed
+- fix: blocking classification is SOTA-only, not for plan-conformance findings
 - fix: settle what the blocked state may do, from the transitions actually made
 - fix: the evaluator could edit the record clearing it
 - fix: repoint repository identity after transfer to digipulse-engineering