Merge commit 'cf0ae6b5a42c35a29437f95cd3104d627038b84d' into HEAD

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

# Conflicts:
#	scripts/harness-version.json
#	test/helpers/agent-sdk-runner.ts

scripts/harness-version.json

@@ -19,7 +19,7 @@
     "scripts/test-paid-shards.ts": {"blob":"5439d0a8f6e1569fa7c924af20d67c30c81d5d30","since":6,"nonBehavioral":"credential rejections are classed infra instead of assertion: no case's pass or fail changes; only failed trials whose provider rejected the run's key move class, so no rerun spends on a rejected key"},
     "scripts/test-pr-profile.ts": {"blob":"44ae88bebf8f716f70a3bd836c8bedb03e6653e8","since":6,"nonBehavioral":"FREE_ONLY_PR_FILES gains scripts/lib/shard-jobs.ts (free-runner module) and .github/workflows/windows-smoke.yml (a workflow no paid case reads); both change only which PR paths select paid cases, never how a case runs or is judged"},
     "scripts/test-strict-output.ts": {"blob":"01185d20dc9761b04b9b2f751ec98664fe0eaf93","since":6},
-    "test/helpers/agent-sdk-runner.ts": {"blob":"ecbd0fd8c7864d09ea7c3cabd106558257eeff3f","since":6,"nonBehavioral":"Cost attribution only (GSTA-150 PR 2): a display-label record carries cost_case_id from runSkillTest's testName and the report binds its cost by it while the failed record and failure class still come from the named records (test/paid-report-cost.test.ts pins both); an aborted Agent SDK session logs its streamed usage as streamed_cost_usd, used only when a case has no record; a listed model-free case with no record or session is a known $0. No case's execution, outcome, failure class, rerun eligibility or panel verdict changes, so no pass-rate series resets"},
+    "test/helpers/agent-sdk-runner.ts": {"blob":"d66dac458b8c13250973e550edf8087d4044cd0b","since":6,"nonBehavioral":"Cost attribution only (GSTA-150 PR 2): a display-label record carries cost_case_id from runSkillTest's testName and the report binds its cost by it while the failed record and failure class still come from the named records (test/paid-report-cost.test.ts pins both); an aborted Agent SDK session logs its streamed usage as streamed_cost_usd, used only when a case has no record; a listed model-free case with no record or session is a known $0. No case's execution, outcome, failure class, rerun eligibility or panel verdict changes, so no pass-rate series resets. Also: runAgentSdkTest now validates systemPrompt before starting; every live caller already passed the claude_code preset, so the prompt the model sees is unchanged and no verdict can change"},
     "test/helpers/auq-substance-panel.ts": {"blob":"4881260ef793668e4963a062db6d06a5e1f7a40f","since":6},
     "test/helpers/claude-pty-runner.ts": {"blob":"ce7a5eaa66e502a052ea9c53348dd2ab25db6148","since":6,"nonBehavioral":"Removes the plan-design 0D focus picker from the shared harness; floor.ts gains an optional pickSetupChoice hook that no other caller passes, so no other case's input or verdict can change. plan-design-finding-floor, the only case whose seed raises that menu, passes the picker from its owned fixture test/fixtures/design-floor-focus-picker.ts, so its owned bytes start its own new series."},
     "test/helpers/codex-host-runner.ts": {"blob":"bff225017e235f0be6ece29f04971ac04b95b544","since":6,"nonBehavioral":"Claude-host verdicts are unchanged: the Codex host (EVALS_HOST=codex, codex-host-runner) runs only when selected and records its own per-model series; the Codex runner refactor (execCodex extraction, onSpawn hook) keeps its events and verdicts and adds token cost accounting; host fields on the plan/report are absent for claude"},

scripts/preflight-agent-sdk.ts

@@ -21,6 +21,7 @@ import '../lib/conductor-env-shim';
 import { query, type SDKMessage } from '@anthropic-ai/claude-agent-sdk';
 import { readOverlay } from './resolvers/model-overlay';
 import { resolveClaudeBinary } from '../lib/claude-bin';
+import { sdkSystemPrompt } from '../test/helpers/sdk-system-prompt';
 
 const USAGE = `Usage: bun run scripts/preflight-agent-sdk.ts [--live]
 
@@ -94,7 +95,7 @@ async function main() {
         prompt: 'say pong',
         options: {
           model: 'claude-opus-4-7',
-          systemPrompt: '',
+          ...sdkSystemPrompt({ type: 'sdk-default', reason: 'connectivity probe with no tools: checks the event stream shape, not skill behavior' }),
           tools: [],
           permissionMode: 'bypassPermissions',
           allowDangerouslySkipPermissions: true,