test(acp): gate the advertised session lifecycle surface in CI

.github/workflows/ci.yml

@@ -290,10 +290,44 @@ jobs:
               retention-days: 7
               overwrite: true
 
+   acp_lifecycle_smoke:
+      if: ${{ always() && !startsWith(github.ref, 'refs/tags/v') && (github.event_name != 'workflow_dispatch' || inputs.rehearsal == 'main-nontag' || inputs.rehearsal == 'nightly-release') && needs.main_plan.outputs.has_tasks == 'true' && needs.main_native.result != 'failure' && needs.main_native.result != 'cancelled' }}
+      needs: [main_plan, main_native]
+      runs-on: ubuntu-22.04
+      timeout-minutes: 15
+      steps:
+         - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5  # v4
+         - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020  # v4
+           with:
+              node-version: "24"
+         - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6  # v2
+           with:
+              bun-version: "1.3.14"
+         - name: Cache bun dependencies
+           uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830  # v4
+           with:
+              path: ~/.bun/install/cache
+              key: bun-1.3.14-${{ runner.os }}-${{ hashFiles('**/bun.lock') }}
+         - run: bun install --frozen-lockfile
+         - name: Download native addon(s)
+           uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093  # v4
+           with:
+              name: main-native-${{ github.run_id }}
+              path: packages/natives/native
+         - name: Run ACP lifecycle smoke
+           # bunfig `[test] pathIgnorePatterns` keeps this out of the default suite, and
+           # naming the path does NOT re-include it: `bun test <path>` filters files that
+           # were already discovered, so a pruned file never matches. Overriding the list
+           # is the only way in. The canonical argv — which restates every repository
+           # ignore pattern so the override does not widen discovery — lives in
+           # ci-dev-affected.ts (BUN_TEST_IGNORE_OVERRIDES / dedicatedTestCommand), so
+           # this job invokes the planner task instead of duplicating the pattern list.
+           run: bun scripts/ci-dev-affected.ts --task=acp-lifecycle-smoke
+
    # Branch protection must keep requiring this stable aggregate status.
    test:
       if: ${{ always() && !startsWith(github.ref, 'refs/tags/v') && (github.event_name != 'workflow_dispatch' || inputs.rehearsal == 'main-nontag' || inputs.rehearsal == 'nightly-release') }}
-      needs: [main_plan, main_native, main_shards, acp_conformance]
+      needs: [main_plan, main_native, main_shards, acp_conformance, acp_lifecycle_smoke]
       runs-on: ubuntu-22.04
       timeout-minutes: 5
       steps:
@@ -303,11 +337,13 @@ jobs:
               native='${{ needs.main_native.result }}'
               shards='${{ needs.main_shards.result }}'
               conformance='${{ needs.acp_conformance.result }}'
-              echo "main_plan=$plan main_native=$native main_shards=$shards acp_conformance=$conformance"
+              lifecycle='${{ needs.acp_lifecycle_smoke.result }}'
+              echo "main_plan=$plan main_native=$native main_shards=$shards acp_conformance=$conformance acp_lifecycle_smoke=$lifecycle"
               test "$plan" = success
               case "$native" in success|skipped) ;; *) echo "native gate failed"; exit 1;; esac
               test "$shards" = success
               test "$conformance" = success
+              test "$lifecycle" = success
 
    nightly_gate:
       if: ${{ always() && (github.event_name == 'schedule' || (github.event_name == 'workflow_dispatch' && inputs.rehearsal == 'nightly-release')) }}

bunfig.toml

@@ -18,6 +18,13 @@ pathIgnorePatterns = [
   "**/node_modules/**",
   ".wt/**",
   ".worktrees/**",
+  # Spawns a broker plus a session host per run and costs tens of seconds, so it
+  # stays out of the default suite. Naming the path on the command line does NOT
+  # re-include it -- `bun test <path>` is a filter over already-discovered files,
+  # so a pruned file can never match. The only way to run it is the canonical
+  # dedicated argv (ci-dev-affected.ts BUN_TEST_IGNORE_OVERRIDES override), which
+  # the planner emits; the fresh-process shard inventory also excludes it.
+  "**/test/acp/acp-lifecycle-smoke.test.ts",
 ]
 
 [run]

packages/coding-agent/CHANGELOG.md

@@ -24,6 +24,7 @@
 ### Added
 - Added implicit discovery and keyless local support for oMLX (`http://127.0.0.1:8080/v1`) with `OMLX_BASE_URL` and `OMLX_API_KEY` configuration.
 
+- ACP session lifecycle (`session/list`, `fork`, `resume`, `close`, `delete`) is now gated in CI. `initialize` advertised all five capabilities to every ACP client and all five worked, but the pinned upstream `acp-core-v1` corpus contains 21 cases and exercised none of them, so the advertised surface had no release-gate coverage. A new stdio smoke test drives the credential-free conformance fixture and asserts the five happy paths, that a forked session id is distinct from its source, that `session/list` returns the created session under `cwd` filtering, and that duplicate `close` is idempotent. Because it spawns a broker plus a session host it is a *dedicated-only* test: `bunfig.toml` prunes it from default `bun test` discovery, the fresh-process shard inventory excludes it (`DEDICATED_ONLY_TESTS`), and exactly one canonical argv — the full ignore-list override from `ci-dev-affected.ts` `BUN_TEST_IGNORE_OVERRIDES` — runs it. The planner routes every path: PR-mode targeted plans emit `test:<path>` with that override, Main CI's full plan emits the `acp-lifecycle-smoke` task, the `acp_lifecycle_smoke` CI job invokes the planner task, and the aggregate `test` job fails closed unless it succeeds. A bunfig prune removes a file from *discovery*, so plain `bun test <path>` — a filter over already-discovered files — can never run a pruned file; any caller that schedules it without the override fails deterministically, which is what both initial CI failures were. Duplicate `close` intentionally pins the already-closed no-op, but the unknown-session *error* shape is deliberately left unasserted: `resume`/`prompt` reject unknown ids with `-32603` while `close`/`delete` no-op on unowned ones (`AcpAgent.closeSession` gates on connection ownership), and whether that asymmetry and that error code are correct is a separately filed open question.
 - Gajae Pet now renders in iTerm2 through a bounded inline GIF protected from ordinary TUI redraws by a reserved raster lease, with the same composer-side layout and lifecycle cleanup guarantees used by Kitty and Sixel.
 - Added the bundled `ouroboros` dark theme, translating the official navy, teal, green, and gold identity into terminal-safe semantic colors while retaining the pet's vivid `#AEE80E` lime and `#7092BE` cool-scale accents. Live previews now recolor the open `/theme` selector and Settings theme submenu instead of leaving their construction-time theme visible, and confirmation consistently settles the preview into the active appearance mapping.
 - Added the 16×16 `Ouroboros` terminal pet: a vivid lime snake with a cool `#7092BE` underside that rests in a soft coil, blinks, flicks its tongue upward, and occasionally sobs with `><` eyes. Its signature flex rolls the same silhouette through a symmetric circle into an exact 180-degree heart pose, blinks a small pink heart twice, and reuses the authored frames in reverse; agent work enters and exits a stable six-frame infinity loop through explicit unwind transitions. Pet skins now own their frame registry, source resolution, idle loop, work transitions, work loop, and signature burst; a saved skin removed by a later installation falls back to RedGajae while an explicit `off` remains off.