| name | opl-runtime-soak-and-recovery-auditor |
| description | Use when auditing OPL runtime/provider environment fit, runtime environment bundles, native helper diagnostics, long-soak evidence, recovery attempts, provider observations, no-regression refs, and route-back decisions without mutating runtime or provider state. |
OPL Runtime Soak And Recovery Auditor
Boundary
Use this skill to audit runtime/provider evidence before Runway, Console, or a release owner consumes it. It is the consolidated reviewer for provider fit, runtime environment bundle proof, native helper diagnostics, long-soak windows, and recovery evidence.
This skill may:
- inspect runtime provider fit, sandbox substrate boundaries, credential risk refs, runtime environment descriptor, materialized root/cache/lock/bundle manifest, reproducibility receipts, native helper/sysprobe/state-index refs, recovery attempts, provider observations, no-regression refs, failure windows, recovery playbooks, and route-back refs;
- classify
provider_fit_gap, credential_boundary_gap, environment_bundle_gap, native_helper_diagnostic_gap, soak_evidence_gap, recovery_gap, provider_observation_gap, no_regression_gap, owner_route_gap, or runtime_ready_overclaim;
- prepare a refs-only audit note that distinguishes observed recovery from readiness.
Optional helper: kernel.py provides deterministic runtime/ref normalization, soak checklist, gap classification, and forbidden-claim lint helpers.
It is stdlib-only, writes nothing, performs no network or subprocess calls, and does not mutate runtime state or claim readiness.
This skill must not:
- submit, wait, harvest, redrive, retry, mutate runtime queues, write provider attempts, manage cloud resources, credentials, endpoints, helper repairs, owner receipts, typed blockers, or readiness ledgers;
- declare runtime ready, provider ready, live ready, production ready, Brand L5, App release ready, or recovery complete;
- treat a clean queue, a single smoke, docs, or AI review as long-soak or recovery authority.
No-authority language: no owner receipts, no typed blockers, no runtime queue writes, no provider attempts, no credential or endpoint lifecycle, no runtime readiness claim, no readiness claims.
Legacy Coverage
This consolidated auditor covers the retired opl-external-runtime-provider-fit-reviewer, opl-runtime-environment-bundle-reviewer, opl-native-helper-diagnostics-reviewer, and opl-runway-recovery-playbook-writer entries. Keep provider fit, environment materialization, helper diagnostics, soak, and recovery evidence together here.
Workflow
- Identify the runtime surface, provider/substrate, environment bundle, helper diagnostic ref, soak window, recovery path, claimed outcome, and consuming owner.
- Group refs by class: provider fit, credential boundary, environment bundle, native helper diagnostic, long-soak, recovery attempt, provider observation, no-regression, failure, owner acceptance, or reviewer note.
- Check whether the refs prove sustained operation, recovery behavior, reproducible environment materialization, helper diagnosis, or only a structural/readback condition.
- Flag missing duration, missing recovery event, stale provider observation, hidden failure, helper authority overclaim, credential risk, environment reproducibility gap, or owner-route ambiguity.
- Recommend the smallest legal route: collect evidence, route to Runway/Console/release owner, route to native-helper owner, or hold the claim.
Output Shape
Return:
verdict: owner_route_candidate, hold, or route_back;
reviewed_refs;
soak_and_recovery_assessment;
runtime_provider_and_environment_assessment;
gaps;
owner_route;
proof_needed;
no_authority_caveat: no owner receipts, no typed blockers, no runtime queue writes, no provider attempts, no runtime readiness claim, no readiness claims.