codex-orchestrator 0.1.44 → 0.1.45

package/docs/deep-dive.md

@@ -391,7 +391,36 @@ use the shared mobile lease described below.
 The proof phase often runs browser automation, such as Playwright, but the same
 contract applies to non-visual proof. A live-smoke proof can exercise an API,
 worker, CLI, or other observable behavior and save the command output as a
-`smoke-output` artifact. The runner provides environment variables for:
+`smoke-output` artifact.
+
+Implementation agents declare the intended proof mode in the scoped completion
+report `proofPlan`. Supported modes are `none`, `non-visual-smoke`, `cli`,
+`api`, `worker`, `browser-visual`, and `mobile-visual`. The child agent must
+choose the narrowest mode that proves the issue; this is an implementation
+intent signal, not proof by itself. The runner remains the authority and
+validates that plan against the issue `Proof Strategy`, changed files, and
+visual/mobile routing signals before publication.
+
+For accepted non-visual modes, the runner uses a report-validation proof path
+instead of dispatching browser or mobile proof. Report validation checks the
+completion report itself:
+
+- `proofPlan.validationCommands` must contain non-empty command strings, and
+  each command must exactly match a passed `validation[].command`;
+- `proofPlan.requiredArtifacts` must contain non-empty artifact targets, and
+  each target must exactly match a reported artifact `path` or `url`;
+- `reviewHandoff.proofByAcceptanceCriteria` must map the evidence to the
+  acceptance criteria.
+
+If those checks pass, the acceptance proof evidence records
+`completion-report:proofPlan` and `completion-report:artifacts` as the proof
+source. If they fail, publication blocks and the next implementation attempt must
+repair the report or provide real evidence. A non-visual `proofPlan` cannot
+downgrade explicit or inferred browser/mobile work: visual/mobile issue strategy
+or changed visual/mobile paths still require visual proof.
+
+For browser, mobile, and adaptive proof phases, the runner provides environment
+variables for:
 
 - issue number;
 - artifact directory;
@@ -431,9 +460,10 @@ dispatch uses one shared policy owner: web/frontend paths route to browser
 proof, while Android, iOS, Flutter, and mobile app paths remain device-backed.
 When acceptance proof is required but the changed paths are backend/API/CLI-only
 and visual proof is not desirable, auto proof does not force a browser or mobile
-target; the runner evaluates the prepared machine-readable
-`acceptance-proof-report.json` and its non-visual artifacts instead. Explicit
-legacy proof command overrides are preserved.
+target; the runner validates the child completion report `proofPlan`, validation
+lines, artifacts, and review handoff instead. Acceptance-proof command selection
+uses only the canonical acceptance-proof policy; `visualProof` command settings
+are not treated as Acceptance Proof compatibility input.
 
 For web UI work, `codex-orchestrator visual-proof browser` reads a proof-owned
 browser scenario, drives Playwright Core against an explicit base URL, and

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codex-orchestrator",
-  "version": "0.1.44",
+  "version": "0.1.45",
   "description": "Reusable GitHub Issues runner for Codex.",
   "type": "module",
   "license": "MIT",

package/prompts/workflows/scoped-implementation.md

@@ -67,6 +67,12 @@ For completed work, include `reviewHandoff` in the runner JSON report:
 
 ## Acceptance Proof
 
+Declare the proof mode in the runner completion report `proofPlan`. Choose the
+narrowest mode that proves the issue: `none`, `non-visual-smoke`, `cli`, `api`,
+`worker`, `browser-visual`, or `mobile-visual`. Do not choose non-visual modes
+for UI or mobile behavior, and do not choose `none` when acceptance criteria need
+observable proof. The runner validates `proofPlan` before publication.
+
 Prepare runner-owned acceptance proof artifacts when configured. For visual work,
 screenshots must be tied back to the acceptance criteria; for non-visual work,
 use smoke outputs, logs, or other observable artifacts. Let the runner execute