Switch Daily Issues Report Generator off the Python sample SDK driver — it ModuleNotFoundErrors on copilot and python3 is absent in the agent container, so the workflow fails 100% of runs.
Problem statement
Daily Issues Report Generator fails on every scheduled run. The Copilot harness is configured to launch the Python sample driver .github/drivers/copilot_sdk_driver_sample_python.py, which imports from copilot import CopilotClient, RuntimeConnection. The agent container has neither the copilot Python package nor a python3 binary, so the driver crashes at import on every retry and the agent never runs.
Evidence
- Latest run §27834964940 (2026-06-19 15:37 UTC,
main, schedule), agent-stdio.log:
[copilot-harness] pre-flight: command not found: python3 (F_OK check failed — binary does not exist at this path)- All 3 attempts (1–3):
ModuleNotFoundError: No module named 'copilot' → attempt N failed: exitCode=1 failureClass=partial_execution
[copilot-harness] done: exitCode=1; agent_output.json = {"items":[]}; TokenUsage=0, Turns=0 (agent never executed).
- Config:
.github/workflows/daily-issues-report.md line 16 → copilot-sdk-driver: .github/drivers/copilot_sdk_driver_sample_python.py.
- Chronic / deterministic: failed 6/6 of the last scheduled runs — 27834964940, 27771124605, 27701060020, 27630903642, 27559871265, 27502926049. Prior run 27771124605 invokes the identical
copilot_harness.cjs python3 .../copilot_sdk_driver_sample_python.py command → same root cause across the series.
Probable root cause
The workflow pins a sample/demo driver intended for environments with the Copilot Python SDK installed. The production agent container ships Node (the default copilot_sdk_driver.cjs works elsewhere in this same window, e.g. SPDD/PR runs) but not python3 nor the copilot Python package, so the Python sample can never start.
Proposed remediation
- Remove the
copilot-sdk-driver override (use the built-in Node driver), or switch to .github/drivers/copilot_sdk_driver_sample_node.cjs.
- If a Python driver is genuinely required, add
python3 + the copilot SDK package to the agent container/runtime setup and gate the workflow on a pre-flight import check.
- Add a compile-time validation that a referenced
copilot-sdk-driver interpreter (python3/node/ruby) is available before the run starts, failing fast with a clear message instead of 3 silent retries.
Success criteria / verification
Daily Issues Report Generator completes a scheduled run with Turns>0 and a non-empty agent_output.json.- Harness no longer logs
command not found: python3 or ModuleNotFoundError: No module named 'copilot'.
- A driver-availability pre-flight check exists and is covered by a test.
Parent: #39883. Related class (Copilot harness failures): #39946 (distinct — that is unclassified exit-1; this is a classified Python import crash).
References:
Generated by 🔍 [aw] Failure Investigator (6h) · 444.2 AIC · ⌖ 12.3 AIC · ⊞ 4.9K · ◷
Switch
Daily Issues Report Generatoroff the Python sample SDK driver — itModuleNotFoundErrors oncopilotand python3 is absent in the agent container, so the workflow fails 100% of runs.Problem statement
Daily Issues Report Generatorfails on every scheduled run. The Copilot harness is configured to launch the Python sample driver.github/drivers/copilot_sdk_driver_sample_python.py, which importsfrom copilot import CopilotClient, RuntimeConnection. The agent container has neither thecopilotPython package nor apython3binary, so the driver crashes at import on every retry and the agent never runs.Evidence
main,schedule),agent-stdio.log:[copilot-harness] pre-flight: command not found: python3 (F_OK check failed — binary does not exist at this path)ModuleNotFoundError: No module named 'copilot'→attempt N failed: exitCode=1 failureClass=partial_execution[copilot-harness] done: exitCode=1;agent_output.json={"items":[]};TokenUsage=0,Turns=0(agent never executed)..github/workflows/daily-issues-report.mdline 16 →copilot-sdk-driver: .github/drivers/copilot_sdk_driver_sample_python.py.copilot_harness.cjs python3 .../copilot_sdk_driver_sample_python.pycommand → same root cause across the series.Probable root cause
The workflow pins a sample/demo driver intended for environments with the Copilot Python SDK installed. The production agent container ships Node (the default
copilot_sdk_driver.cjsworks elsewhere in this same window, e.g. SPDD/PR runs) but notpython3nor thecopilotPython package, so the Python sample can never start.Proposed remediation
copilot-sdk-driveroverride (use the built-in Node driver), or switch to.github/drivers/copilot_sdk_driver_sample_node.cjs.python3+ thecopilotSDK package to the agent container/runtime setup and gate the workflow on a pre-flight import check.copilot-sdk-driverinterpreter (python3/node/ruby) is available before the run starts, failing fast with a clear message instead of 3 silent retries.Success criteria / verification
Daily Issues Report Generatorcompletes a scheduled run withTurns>0and a non-emptyagent_output.json.command not found: python3orModuleNotFoundError: No module named 'copilot'.Parent: #39883. Related class (Copilot harness failures): #39946 (distinct — that is unclassified exit-1; this is a classified Python import crash).
References:
Related to [aw-failures] [aw] Failure Investigation Report — 6h window (2026-06-17 19:34 UTC) #39883