[aw-failures] Smoke Copilot Sub Agents 100% red — Copilot SDK returns "No model available (policy enablement)" for gpt-5.3-codex [Content truncated due to length] #42824

Fix Smoke Copilot Sub Agents first — it has never once passed and burns full retries on every run.

Every run of Smoke Copilot Sub Agents since the workflow was first triggered today has hard-failed with the Copilot SDK driver reporting No model available. Check policy enablement under GitHub Settings > Copilot for model gpt-5.3-codex. The harness does not recognize this as a model/policy error, so it exhausts all 3 retries and emits an unclassified hard-red. Zero tokens are ever consumed — the model is never reached.

Failure clusters (last 6h)

Evidence — C1 (representative run §28534423464)

[copilot-sdk-driver] [sdk-driver] creating session with model="gpt-5.3-codex" providers=1 models=32
[copilot-sdk-driver] [sdk-driver] session created: sessionId=...
[copilot-sdk-driver] [sdk-driver] sending prompt...
[copilot-sdk-driver] [sdk-driver] error: Execution failed: Error: No model available. Check policy enablement under GitHub Settings > Copilot
[copilot-harness] attempt 1 failed: exitCode=1 failureClass=partial_execution isModelNotSupportedError=false isMCPPolicyError=false isAuthError=false hasOutput=true retriesRemaining=3
[copilot-harness] attempt 2 failed: exitCode=1 ... retriesRemaining=2
[copilot-harness] attempt 3 failed: exitCode=1 ... retriesRemaining=1

• 100% red, never green: all 4 runs of this workflow (first at 06:30Z today) failed — this is not a regression, the workflow has never succeeded.
• Stable, not drift: audit-diff of §28519811362 vs §28534423464 shows 0 token usage in both, no firewall anomalies, no domain/status drift. The model is never reached; the failure is deterministic.
• Misclassification: the driver error string clearly indicates model/policy unavailability, yet isModelNotSupportedError=false and isMCPPolicyError=false, so the harness treats it as partial_execution and retries 3× pointlessly before failing.

Existing-issue correlation

• C2 (Smoke CI EACCES firewall logs) is already tracked by [aw-failures] Smoke CI hard-red at startup — EACCES mkdir /tmp/gh-aw/sandbox/firewall/logs, agent never invoked (rootless left [Content truncated due to length] #42398 — that issue is still actively firing (latest at 18:10Z) so it stays open; no new issue needed. Note Smoke CI is only intermittently red (successes at 18:30Z/19:07Z), unlike C1.
• C1 matched no open agentic-workflows issue (searched No model available, policy enablement, Sub Agents) → genuine coverage gap.
• No open tracking issue showed fresh evidence of being fixed this window, so none were closed.

Fix roadmap

P0 — unblock Smoke Copilot Sub Agents (pick one, in order):

1. Confirm gpt-5.3-codex is enabled in the org/account Copilot policy used by this workflow; if it is intentionally unavailable, repin the workflow to an enabled Copilot model.
2. In the Copilot harness, classify No model available. Check policy enablement as a model-not-supported / policy error (set isModelNotSupportedError/isMCPPolicyError) so it fails fast with a clear message instead of retrying 3×.

P1 — Smoke CI firewall-logs EACCES: tracked in #42398, no duplication here.

P2 — Auto-Triage cancellations: no action (runs were cancelled, not failed).

Success criteria

• Smoke Copilot Sub Agents produces at least one green run (agent reaches a model, tokens > 0).
• The Copilot harness surfaces No model available as a recognized model/policy failure and stops retrying it 3×.

References: §28534423464 · §28519811362 · §28538082921

Generated by 🔍 [aw] Failure Investigator (6h) · 199 AIC · ⌖ 31.5 AIC · ⊞ 5.2K · ◷

• expires on Jul 8, 2026, 11:33 AM UTC-08:00