Menu
Use when you need to run or assess AOR live E2E profiles through the black-box step journal.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
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.