KB-5E04

Codex Agent Data Session Stability Report 2026-03-23

3 min read Revision 1
agent-datasession-stabilitycodexreport2026-03-23

Codex Agent Data Session Stability Report — 2026-03-23

Scope

  • Repo: agent-data-test only
  • Patch scope: thin / unchanged
  • No changes to web-test
  • No runtime gate added to /chat
  • No runtime gate added to MCP tool-call chung
  • No Directus system_issues upsert

Patch Summary

  • Added agent_data/session_readiness.py
    • health gate
    • sentinel query gate
    • retry/backoff 0s -> 2s -> 5s -> 10s
    • error classification: backend_down, tool_route_down, session_binding_failed
    • cache for repeated bootstrap checks in same session
  • Added POST /session-ready in agent_data/server.py
  • Added stdio bootstrap gate in mcp_server/stdio_server.py
    • calls /session-ready once per stdio process/session
  • Added self-test CLI scripts/test-agent-data-session-ready.sh
  • Added offline tests:
    • tests/test_session_readiness.py
    • tests/test_stdio_session_gate.py
    • updates in tests/test_server.py
    • helper stub tests/langroid_test_stubs.py

Local Verification

Commands run:

venv/bin/python -m py_compile agent_data/session_readiness.py agent_data/server.py mcp_server/stdio_server.py
bash -n scripts/test-agent-data-session-ready.sh
venv/bin/python -m pytest -o addopts='' tests/test_session_readiness.py tests/test_stdio_session_gate.py tests/test_server.py -q

Results:

  • py_compile: PASS
  • bash -n: PASS
  • pytest: PASS (35 passed in 1.41s)

PR / CI State

External Blocker

GitHub check annotations show the same external blocker on multiple jobs (guard, lint, Parse Test Suite):

The job was not started because recent account payments have failed or your spending limit needs to be increased. Please check the 'Billing & plans' section in your settings

Evidence collected via GitHub API:

  • Check run 68155654723 (guard)
  • Check run 68155654677 (lint)
  • Check run 68155654646 (Parse Test Suite)

This means:

  • CI jobs are not failing because of the patch logic
  • CI jobs are not starting because of GitHub billing/account state
  • Merge cannot be completed through normal protected-check flow until billing is restored

Repo Boundary / Safety

  • Repo touched: agent-data-test only
  • No web-test approval needed
  • No agent URL/config drift introduced
  • No runtime behavior change on shared /chat or generic MCP tool-call path
  • Change is limited to explicit readiness endpoint + stdio bootstrap + self-test + offline tests

Kết luận

  • Patch scope: thin / unchanged
  • Tests run: py_compile, bash -n, targeted offline pytest (tests/test_session_readiness.py, tests/test_stdio_session_gate.py, tests/test_server.py)
  • Tests pass/fail: local PASS; GitHub CI jobs not started
  • Root blocker còn lại: GitHub Actions billing/account issue prevents required jobs from starting
  • Repo touched: agent-data-test only
  • safe_to_merge: NO
  • Final action: blocked