fix(agents): disable pi-coding-agent auto-retry to prevent tool call replay loops (#84798)

Summary:
- The PR disables pi-coding-agent auto-retry inside prepared embedded Pi settings, updates the focused settings test, and moves the changelog entry into Unreleased.
- Reproducibility: yes. source-reproducible: current main leaves embedded Pi retry enabled, while pi-coding-ag ... e assistant error before continuing. I did not run a live Feishu/Qwen replay loop in this read-only review.

Automerge notes:
- PR branch already contained follow-up commit before automerge: fix(agents): disable pi-coding-agent auto-retry to prevent tool call …

Validation:
- ClawSweeper review passed for head ca745fd55d.
- Required merge gates passed before the squash merge.

Prepared head SHA: ca745fd55d
Review: https://github.com/openclaw/openclaw/pull/84798#issuecomment-4504702875

Co-authored-by: yelog <yelogeek@gmail.com>
Co-authored-by: clawsweeper <274271284+clawsweeper[bot]@users.noreply.github.com>
Co-authored-by: clawsweeper[bot] <274271284+clawsweeper[bot]@users.noreply.github.com>
Approved-by: takhoffman
Co-authored-by: takhoffman <781889+takhoffman@users.noreply.github.com>

CHANGELOG.md

@@ -13,6 +13,7 @@ Docs: https://docs.openclaw.ai
 - Media/audio: skip empty structured sherpa-onnx transcripts instead of treating the raw JSON payload as spoken text. (#84667) Thanks @TurboTheTurtle.
 - Memory-core/dreaming: reuse stable narrative subagent session keys per workspace and phase while keeping per-run idempotency and bounded cleanup, so stale `dreaming-narrative-*` sessions do not accumulate. Fixes #68252, #69187, and #70402. (#70464) Thanks @chiyouYCH.
 - Trajectory/support: tolerate partial skill snapshot entries when building support metadata so rejected skill path scans no longer abort trajectory capture. (#71185) Thanks @lukeboyett.
+- Agents/Pi: disable the embedded pi-coding-agent runtime auto-retry so OpenClaw's own retry and failover loop does not replay failed tool calls through a nested SDK retry. Fixes #73781. (#74434) Thanks @yelog.
 - CLI/perf: keep `setup --help`, `onboard --help`, and `configure --help` out of the full wizard runtime while preserving the existing help output. (#84488) Thanks @frankekn.
 - CLI/perf: keep `agents --help` out of agents action/runtime imports so help, completion, and command discovery paths avoid loading the full agents runtime. (#84483) Thanks @frankekn.
 

src/agents/pi-project-settings.test.ts

@@ -161,9 +161,8 @@ describe("createPreparedEmbeddedPiSettingsManager", () => {
       });
 
       expect(settingsManager.getShellCommandPrefix()).toBe("echo trusted &&");
-      expect(settingsManager.getRetryEnabled()).toBe(true);
+      expect(settingsManager.getRetryEnabled()).toBe(false);
 
-      settingsManager.setRetryEnabled(false);
       await settingsManager.flush();
 
       const diskSettings = JSON.parse(await fs.readFile(agentSettingsPath, "utf8")) as {

src/agents/pi-project-settings.ts

@@ -61,5 +61,10 @@ export function createPreparedEmbeddedPiSettingsManager(params: {
     cfg: params.cfg,
     contextTokenBudget: params.contextTokenBudget,
   });
+  // Disable the pi-coding-agent auto-retry. OpenClaw has its own comprehensive
+  // retry layer (failover rotation, auth profile rotation, empty-error retry,
+  // thinking-level fallback) in run.ts. Having both layers active creates a
+  // double-retry that can replay failed tool calls in an unbounded loop (#73781).
+  settingsManager.setRetryEnabled(false);
   return settingsManager;
 }