Skip to content

bug: adversarial checkpoint snapshot is orphaned from artifact manifest #2179

Description

@chubes4

Severity

Medium: a successful adversarial run produces a bundle that fails WP Codebox's own artifact verifier.

Environment

  • WP Codebox 0.18.0
  • Playground adapter 0.18.0
  • Playground CLI 3.1.46
  • Run ID: run_35552ed2c2c241198cd354aee9ce8c30
  • Runtime ID: runtime-ms9e8oaf-er7o0u
  • Seed: community-impossible-oracle-control-260

Reproduction

  1. Run a small adversarial campaign with resetPolicy.mode=checkpoint-per-case and replay selection findings.
  2. Confirm the campaign succeeds and writes a finding/replay.
  3. Run:
wp-codebox artifacts verify --bundle <runtime-bundle> --json

Expected

The successful bundle verifies, and every generated checkpoint/snapshot file is represented in manifest.json.

Actual

Verification fails:

{
  "valid": false,
  "violations": [{
    "code": "orphaned-file",
    "path": "files/runtime-snapshots/snapshot-ms9e8ygz-29gn6o.json",
    "message": "Bundle file is not listed in manifest.json"
  }]
}

Impact

  • Breaks artifact integrity guarantees for checkpoint-backed campaigns.
  • Makes an otherwise valid oracle-sensitivity control bundle unsuitable for transfer/review.
  • Replay exists, but the parent evidence bundle cannot be trusted as verifier-clean.

Evidence

  • Bundle: /mnt/extrachill-workspace/tmp/opencode/community-posting-fuzz-cc5e8b0-control-artifacts/runtime-ms9e8oaf-er7o0u
  • Campaign result: files/adversarial/impossible-oracle-control/result/adversarial-campaign-result.json
  • Replay: files/adversarial/impossible-oracle-control/replay/6c23a4b07676fbb794bcd7a9c688748747ef70ae98ca79841b0ae59c752959d9.json
  • Orphan: files/runtime-snapshots/snapshot-ms9e8ygz-29gn6o.json

Suggested ownership layer

WP Codebox artifact manifest finalization for runtime checkpoint/snapshot files.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions