jleechan2015

churn-guard.sh — Reproduction Steps

Setup

git clone https://github.com/jleechanorg/agent-orchestrator.git
cd agent-orchestrator
git checkout feat/churn-guard

Test 1: Non-PR commands pass through

jleechan2015

PR 383 repro: git fetch && git checkout chore/claude-md-fork-reference && pnpm install && pnpm --filter @jleechanorg/ao-core test

jleechan2015

Evidence: rewards_pending_until_level_up_real_e2e (MCP + real LLM)

Self-contained summary aligned with evidence-standards (real execution, not mock; git + server provenance in metadata.json; Claim → Artifact Map in evidence.md).

1. What was proven

Claim	Proof artifact	Key fields
After GOD_MODE_UPDATE_STATE with inconsistent rewards_pending (processed: true while level-up still owed), server clears stale processed	scenario_results_checkpoint.json	rewards_pending_after_god.processed → false
Same after one real LLM story turn	same file	rewards_pending_after_llm.processed → false

jleechan2015

Repro — bd-elcfg (PR #393)

Config-only change to agent-orchestrator.yaml.example.

./scripts/validate-config.sh agent-orchestrator.yaml.example
pnpm build
AO_WHOLESOME_PR_TITLE='[agento] feat: enable evolve loop for agent-orchestrator in yaml example' pnpm test

jleechan2015

╔═══════════════════════════════════════════════════════╗
║  Fix CharacterCreationAgent hijacking LevelUpAgent    ║
║  Recorded: 2026-04-07T07:11:32Z                 ║
╚═══════════════════════════════════════════════════════╝

━━━━━━━ 1. GIT PROVENANCE ━━━━━━━
  HEAD SHA:  3b49337d0e7e023a077cf64124db9963fdf82baf
  Branch:    fix/level-up-agent-hijack
  Merge-base vs main: e089c2539b3aa867b236975c56817dac3def778f
  Commits ahead of main: 4

jleechan2015

{"version": 2, "width": 80, "height": 24, "timestamp": 1775546341, "env": {"SHELL": "/usr/bin/bash", "TERM": "xterm-256color"}}
[0.008326, "o", "no sessions\r\n"]

jleechan2015

mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_when_rewards_box_available_but_rewards_pending_null PASSED [ 26%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_with_multilevel_target_from_xp PASSED [ 31%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_ignored_when_rewards_box_stale PASSED [ 36%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_rewards_box_false_no_injection PASSED [ 42%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_rewards_box_none_no_injection PASSED [ 47%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choices_both_rewards_pending_and_rewards_box_set PASSED [ 52%]
mvp_site/tests/test_level_up_stale_flags.py::TestRewardBoxLevelUpInjection::test_inject_levelup_choi

jleechan2015

PR #6138 Evidence Bundle

Git Provenance

• HEAD SHA: 3b49337d0e7e023a077cf64124db9963fdf82baf
• Branch: fix/level-up-agent-hijack
• Merge-base vs main: e089c2539b3aa867b236975c56817dac3def778f
• Commits ahead of main: 4

Commit Log

jleechan2015

Main Branch Sync Evidence

Git Provenance

• HEAD SHA: afdaff4c466a0d7857b8588d25cd7fca37c44343
• Branch: worktree_level2
• Merge-base vs main: afdaff4c466a0d7857b8588d25cd7fca37c44343 (synced)
• Commits ahead of main: 0 (synced)

Recent Main Commits (from origin)

jleechan2015

PR #6138 Evidence Review - UPDATED

Original Review (Before Test Run)

• Verdict: FAIL (missing evidence bundle)
• Confidence: LOW

After Running E2E Test

New Evidence Bundle

Location: /tmp/worktree_levelup_fix/fix_level-up-agent-hijack/level_up_agent_hijack_real_e2e/iteration_003/