anvil-serving-workbench

name	anvil-serving-workbench
description	Operate anvil-serving from OpenClaw using MCP/controller-backed playbooks, CLI fallbacks, and sub-agent evidence packets. Use for router status, model selection, preflight, benchmark probes, OpenClaw sync, promotion evidence, and safe small-model workflow slices.

Anvil Serving Workbench

Use this skill as the OpenClaw-visible entry point for operating anvil-serving. It mirrors the repo workbench skill used by Codex and Claude Code.

Rules

• List the current MCP catalog first. Use operation_contracts before topology-aware or controller-backed work so the selected operation, target context, transport, and MCP wrapper agree. Use the complete grouped map below instead of relying on a memorized subset. If the MCP server/controller is unavailable, report that loss of coverage explicitly.
• Use documented anvil-serving CLI verbs only when a structured tool is missing. Safe fallbacks are read-only or preview-first verbs such as eval usage, models sync, models recipes, models score, harness sync openclaw --out -, and other render/inspect commands. Return the command preview as evidence and name the missing MCP wrapper as a product gap. Before using a CLI fallback, verify that it resolves to the intended checkout or installed version; do not imply it is controller-backed.
• Use 127.0.0.1 for local URLs. Do not use localhost.
• For resource-owned commands, pass the deployed topology and declare the actual command host/runtime; do not infer command identity from the target.
• For OpenClaw Talk / Anvil Voice work, record the command host before interpreting loopback. Reference deployment: Fakoli Mini runs OpenClaw Gateway plus Anvil Voice Realtime/proxy and reserves its 16 GB RAM for OpenClaw, Claude Code, and Codex. Do not run STT/TTS/LLM model serves on Mini for reference testing. Fakoli Dark owns the router, candidate serves, and STT/TTS model endpoints. Prefer dark-audio or mini-dark-audio-proxy for OpenClaw Talk; mini-audio is an explicit optional same-host/local-audio mode only. A non-gateway checkout failing to reach Mini proxy loopback is topology evidence, not proof the live Mini/Dark path is down.
• Pass credentials by environment variable name only.
• Stop for a human gate before profile promotion, router policy changes, metered cloud enablement, destructive cache/host repair, Docker/WSL restart, or public/non-loopback bind.
• Treat host_summary, gpu_inventory, and cache_prune_plan as read-only. Use host_manage only for an exact reviewed repair after a human gate. Report Docker/WSL restart, WSL config edits, and cache deletion as blocked or human_required unless the human approves the existing CLI gate.
• Use router_promote with validate_only=true, dry_run=false, and confirm=true for bounded, network-isolated, already-local selected-image validation, or the default dry_run=true for preview. Live apply requires confirm=true, dry_run=false, and human_approved=true. Keep promoted=false unless a human-approved promotion actually ran.
• Treat serves_promote as preview/validation unless confirm=true, dry_run=false, and human_approved=true are present. Mutating router_transition actions are also preview-first and human-gated.
• Never let the model being evaluated grade its own output.

MCP Tool Map

• Router: router_status, router_logs, router_manage, router_transition, decision_summary, route_decision, and router_promote.
• Serves and residency: serves_status, reservation_status, serves_manage, serves_logs, and serves_promote.
• Voice: voice_manage and voice_proxy_manage.
• Host, models, and telemetry: doctor_summary, host_summary, gpu_inventory, observability_collect, host_manage, models_inventory, and cache_prune_plan.
• Harness: openclaw_sync, openclaw_gateway_status, and openclaw_gateway_restart.
• Evaluation and evidence: preflight_probe, benchmark_probe, benchmark_artifact, workflow_packet_validate, external_bench_sources, external_bench_list, external_bench_report, and external_bench_compare.
• Transport discovery: operation_contracts.

benchmark_probe and benchmark_artifact are bounded capacity tools, not quality graders. Repeated quality evaluation remains the CLI-only anvil-serving eval benchmark quality workflow until a dedicated wrapper exists.

Playbooks

• Readiness: inspect operation_contracts, router, serves, reservation_status, doctor, host/GPU, model inventory, gateway status, and configured endpoint status. Use observability_collect only for bounded declared capabilities.
• Model catalog: read or sync model inventory and use external_bench_sources, external_bench_list, external_bench_report, or external_bench_compare for benchmark priors. Keep those priors advisory-only. Inspect recorded configurations with models recipes list/show. Create or revise candidates with models recipes create/update, review the rendered recipe, and run models recipes load only with an exact container and confirmation. Use models recipes delete only for an exact reviewed registry entry. Gate models pull on its network, disk, and target volume.
• Serve swap: inspect reservation_status, preview lifecycle work with serves_manage or named transactions with serves_promote, and inspect bounded logs. The role-based recipe flow is anvil-serving serves switch ROLE [MODEL]; it is CLI-only, so return its preview and name the missing MCP wrapper. Require an exact target and the documented confirmation gate, then run preflight before benchmarking.
• Voice lifecycle: preview audio with voice_manage and the persistent Realtime proxy with voice_proxy_manage; live lifecycle changes on the owning host require confirm=true plus dry_run=false. Profile selection is topology selection: for reference OpenClaw Talk and candidate benchmarks, keep Mini model-free and select Dark-host audio or a Mini-side proxy to Dark. Use mini-audio only when explicitly testing the optional same-host/local-audio mode.
• Harness sync: preview OpenClaw provider, skill, and agent config with openclaw_sync; apply only to an explicit out or gateway_host target.
• Router operations: use bounded status/log/decision tools and router_transition; mutations remain preview-first.
• Evaluation: use preflight_probe with explicit thinking mode, reasoning effort/evidence, visible-answer budget, reasoning headroom, and allowed finish reasons. Use the MCP benchmark tools for capacity only. Quality runs must use repeated eval benchmark quality attempts and preserve visible output, reasoning-channel evidence, finish reasons, separate budgets, provenance, and per-attempt failure classification.
• Promotion evidence: assemble status, decision summaries, route probes, preflight, benchmark artifacts, calibration, profile/config diffs, and reviewer recommendation with promoted=false.
• OpenClaw COLO smoke/eval: for Mini-to-Dark validation or release/blog stats, use examples/openclaw/colo_smoke.py --live; add --run-generations --run-interaction-benchmark for repeatable intent stats. Preserve interaction_benchmarks and recipe fields. Treat generation caps, exact/stream benchmark caps, benchmark reasoning effort, and per-intent overrides as router tier params owned by the model recipe, not as skill or plugin constants.
• Host/cache work: collect host_summary and cache_prune_plan; MCP cache pruning is plan-only and must not delete. host_summary also reports the persistent host.toml WSL cache-reclaim policy. When enabled, confirmed model pulls and managed model-load boundaries inherit one disclosed, readiness-aware, warning-only page-cache postcondition; do not add a second manual reclaim after a reclaimed result. Ad-hoc Compose, voice, request-time ComfyUI loads, and the request-triggered multiplexer remain outside that policy. Treat changing the persistent opt-in as a reviewed host mutation.

Model Benchmark Source Freshness

When researching Fast or Heavy tier candidates, include dates in the evidence. Treat serving posts, community recipes, benchmark tables, and issue threads as time-sensitive.

• Classify sources as current at 0-60 days, aging at 61-120 days, and stale after 120 days, when undated, or when tied to a materially older engine, driver, CUDA, quantization, or checkpoint generation.
• Prefer current official model cards, vendor release notes, and vLLM, SGLang, llama.cpp, NVIDIA, or model-owner documentation and issues.
• Treat Reddit, forum posts, and community benchmarks as recipe discovery, not promotion evidence. Label aging or stale hardware-matched posts as historical priors and require a current official source or local benchmark.
• Record candidate, URL, published or observed date, age class, evidence type, hardware/engine relevance, and decision impact.
• Never let stale community evidence justify a shortlist, serve swap, or promotion by itself. Local preflight, benchmarks, independent quality eval, and the human gate remain required.

Roles

Use small/local models for inventory scout, route analyst, preflight runner, benchmark runner, and evidence reporter. Use a stronger independent critic for promotion recommendations and adversarial review.

Packet

Return schema_version: "operator-workflow/v1", request, gate_state, targets, tools_used, artifacts, advisory_priors, recommendation, human_gate_required, and promoted. Each tools_used entry must include source_class, ok, dry_run, confirmed, target, and error. Allowed gate_state values are not_required, confirm_required, human_required, and blocked. Allowed recommendations are promote, do_not_promote, needs_more_data, and blocked. Validate final packets with workflow_packet_validate when the MCP/control-plane tool is available. That validator checks packet shape, gate consistency, evidence scope, and bounded paths; it does not prove evidence sufficiency or reviewer independence. An independent critic must still reject promotion when the evidence is missing or self-generated.