Menu
Use when you need to run or assess AOR live E2E profiles through the black-box step journal.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Use when you need to choose a delivery slice, derive local tasks, split a slice, or verify that a change still matches the current roadmap.
Use when you need to prepare, review, or update a live E2E rehearsal profile for AOR.
Use when preparing, reviewing, validating, or publishing an AOR npm CLI alpha release for @grinrus/aor.
Use when a feature or flow change adds fields, changes packets, changes reports, or affects routes, wrappers, policies, or APIs.
Use when you need to decide where a change belongs in the AOR repository or which docs, examples, and modules must change together.
Use when you need to verify that a change maps to supported user stories or when you need to check that a workflow still closes the intended role-based outcome.
| name | live-e2e-runner |
| description | Use when you need to run or assess AOR live E2E profiles through the black-box step journal. |
docs/ops/live-e2e-standard-runner.md and docs/ops/live-e2e-target-catalog.md.live_e2e.flow_range_policy, live_e2e.installation_policy, live_e2e.interaction_capability, live_e2e.frontend_capability, live_e2e.safety_policy, operator policy fields, and implementation_loop.node ./scripts/live-e2e/manual-live-e2e.mjs --project-ref . --profile <profile> --run-id <id>agent_decision_request_ref;--operator-decision-file <decision.json> when the step can continue;requested_interaction through aor run answer or the HTTP answer route;run-id after completing any required public action.run-profile.mjs only as the strict proof orchestrator for skill-agent-only profiles. It must fail closed when operator decisions are missing or rejected:
node ./scripts/live-e2e/run-profile.mjs --project-ref . --profile <profile>node ./scripts/live-e2e/step-evaluator.mjs --project-ref . --profile <profile>node ./scripts/live-e2e/qualification-loop.mjs --project-ref . --profile <medium-or-larger-profile>passed means the run can count toward provider qualification;needs_fix means inspect run-health gaps in the analysis artifact, patch AOR or the run setup, commit, and rerun from a fresh isolated workspace;blocked means fix credentials, provider setup, safety, or environment before judging product quality.step_journal[].plan;agent_decision_request_ref, live_e2e_step_observation_files[], command transcripts, artifact refs, UI/API/log output, and live_e2e_controller_state_file;aor_installation_proof_file and setup_journal[] before trusting SDLC step evidence;continue when deterministic and semantic analysis pass;answer when requested_interaction is present;frontend_interact when the profile exposes a guided web surface and the step needs browser evidence;retry_public_step only through public CLI/API/web surfaces;diagnose when evidence is incomplete or quality failed;block when safety, policy, or non-resumable continuation prevents progress.aor run answer or the HTTP answer route). Verify the answer audit ref, state_history[], and final interaction_status before continuing.frontend_interactions[] and the linked web smoke evidence. AOR operator browser/UI evidence must be tied to the relevant step observation. When the profile requires browser-task-proof, use the app_url from guided_browser_task_proof_request_file as the live browser target and treat smoke_app_url only as short-lived render-guardrail metadata. The request should carry expected proof, HTML, DOM, accessibility, visual guardrail refs, and required_accessibility_checks[]; after writing proof, ensure manual-live-e2e.mjs --prepare-decision hydrates the proof JSON/screenshots/accessibility checks into inspected AOR operator UI refs and final observation/run-health use the hydrated proof refs. Missing proof, missing screenshot/visual evidence, or missing structured accessibility check refs is a factual run-health ui_validation blocker; do not treat deterministic aor app --smoke refs as sufficient UX proof.execution#N -> review#N repair loop when needed, eval, delivery, and profile-enabled release/learning.live_e2e_run_health_report_file before discussing outcome quality. Read diagnostic_health separately from primary run completion: optional diagnostic timeouts/failures are factual run warnings or failures, not outcome-quality verdicts. Separate run-health failures (owner, phase, class, command/controller/provider/target/environment/operator/evidence gaps) from code, artifact, delivery, UI/UX, accessibility, and traceability quality.running, silent-running, and timeout-risk provider states as transient. Do not diagnose or block only from stale observation evidence if durable run-control state has a newer terminal provider status. For compiled_context_budget_exceeded, inspect request_artifact_ref, provider_work_packet_ref, context_budget_status, top_context_size_sources[], and raw adapter evidence before planning a fix.implementation_repair_loop_exhausted and review_quality_not_approved as factual run-health blockers from final review evidence. Do not continue into QA or prepare an outcome quality assessment for that run.node ./scripts/live-e2e/quality-assessment.mjs prepare --run-summary-file <live_e2e_run_summary_file>--paired-aor-operator-ui-run-summary-file <guided_summary> to prepare;live-e2e-quality-assessment-report as the SWE evaluator in free-form expert mode using the structured dimensions and findings taxonomy;node ./scripts/live-e2e/quality-assessment.mjs validate --assessment-report-file <report>.node ./scripts/live-e2e/quality-assessment.mjs gate --policy all-pass --assessment-report-file <report> after validation. Treat failures as local fix-and-rerun findings only; do not mutate run-health or provider qualification from the quality gate.live_e2e_run_health_overall_status, live_e2e_run_health_report_file, implementation_loop.iterations[], controller state, AOR operator UI/UX evidence when present, quality assessment status/gap report/gate status when available, and any step journal findings. Do not report legacy runner_quality_summary, quality_judgement, canonical_status, final_skill_agent_verdict_file, agent_operator_assessment, verdict_matrix, step_matrix, artifact-quality matrix fields, or path-scope verdicts.