5
Live HTTP surfaces
Paper 1.17 · Operator Shell Pattern
OpenClaw Wrapper
OpenClaw is the operator-facing wrapper around Project Phoenix: the shell that makes deterministic backends accessible without becoming the authority.
7
Completed wrapper use cases
4
Tracked incidents
1
Validated behavioral contracts
0
Correctness decisions delegated to model
Working rule: OpenClaw makes Project Phoenix accessible. Project Phoenix makes OpenClaw outputs trustworthy.
The Claim
Project Phoenix already had deterministic authority: grounded domain tools, solver-backed computation, and benchmark-backed evaluation. What it lacked was an operator-facing wrapper that exposed those capabilities without forcing the operator to know the internal repo structure.
OpenClaw fills that gap as an outer layer for access, compression, monitoring, and workflow discipline. The wrapper is useful only if the authority boundary holds. The shell can expose, summarize, and verify. It cannot become the place where correctness is decided.
Three-Tier Architecture
Shell
OpenClaw
HTTP surfaces, operator packets, hardening gate, and incident workflow.
↓
Compression
ShowcaseAgent
Routing and domain compression layer. Exposes Phoenix findings in operator-readable form.
↓
Authority
Phoenix Backends
Deterministic domains and solvers. Correctness, grounding, and benchmark truth stay here.
Access
Expose deterministic backends through stable endpoints and scripts instead of tribal knowledge.
Compression
Turn raw JSON, traces, and CSV artifacts into operator-readable packets and decision surfaces.
Discipline
Enforce incident handling, hardening checks, and documentation review so workflow decay is visible.
Live Operator Surfaces
| Surface | Role | Current use |
|---|---|---|
/phoenix-ops-summary |
Authority snapshot | Compressed system state across Phoenix backends. |
/phoenix-ops-status |
Backend health | Per-backend health, aggregate checks, and recent invocation log. |
/phoenix-ops-workspace |
Operator workspace | Starting surface for status, routing, screenshot, and incident context. |
/phoenix-ops-trends |
Trend reporting | Compact rolling view of latency, health counts, and summary history. |
/phoenix-benchmark-summary |
Benchmark packet | Deterministic benchmark exposure through the wrapper. |
Hardening Gate
scripts/check_openclaw_hardening.sh is the single verification command for endpoint health, threshold compliance, and artifact state.
Modes: default live check, --record for milestone audit runs, --skip-live for artifact-only validation.
Incident Workflow
openclaw_repair_packet.sh through openclaw_validate_incidents.sh tracks operator-visible failures from capture to resolution.
Resolution behavior is now governed by a validated PBC behavioral contract (scripts/pbc/openclaw-resolution.pbc.md), schema-conformant and regression-testable. The contract has documentation authority; enforcement gate is the next step.
Seven Use Cases
Paper 1.17 selected use cases under one rule: do not add features in search of use cases; add use cases that prove the wrapper is useful.
Measurement
1. Benchmark Review
openclaw_benchmark_review.sh
Compresses benchmark artifacts into an operator packet. Makes harness importance and protocol-tier results visible without manual JSON inspection.
2. Run-Trace Triage
openclaw_run_trace_triage.sh
Maps recurring failure families to replay targets and escalation verdicts. Separates live design gaps from legacy capture noise.
4. Model Comparison Packet
openclaw_model_comparison_packet.sh
Packages local model comparisons with transport-validity metadata. Encodes the Paper 1.16 capture-integrity correction.
Operator Discipline
3. Incident Repair Loop
openclaw_repair_packet.sh
Tracks real breakages through before-state capture, fix, after-state evidence, and validation. One stale run-trace script was fixed end-to-end through this loop.
5. Documentation Status Review
openclaw_doc_status_review.sh
Runs deterministic drift checks across operator documents. Found and cleared stale claims, an outdated current-state note, and missing script references.
Decision Surfaces And Demos
6. Routing Policy
openclaw_routing_policy.sh
Encodes a four-lane routing policy backed by benchmark evidence: deterministic, local repair-assisted, local strict protocol, and strong-model API.
7. Solver-Backed Demo
openclaw_showcase_routing_demo.sh
Confirms the same pattern seen in TSP: rule routing stays authoritative, while forced LLM routing introduces the failures.
The Boundary That Did Not Move
What the wrapper does
- Expose deterministic backends through stable operator surfaces
- Summarize artifacts into readable packets and dashboards
- Enforce verification, incident handling, and documentation hygiene
What the wrapper does not do
- Decide correctness at the endpoint level
- Bypass Phoenix in the name of convenience
- Use model judgment where deterministic outputs already exist
OpenClaw never became the authority layer. The shell stayed outside. The authority stayed inside. Correctness stayed in the deterministic layer.
Current Limits
Sandbox visibility
Sandboxed localhost checks do not always see the live gateway. Some endpoint verification still needs the live shell environment.
Open routing gaps
Stan regularization and ParableAgent teaching-complexity gaps remain tracked as incidents and still need routing-surface fixes.
Clean-capture reruns
Pipeline and handoff trace lanes still need post-fix reruns under clean ollama_api capture.
Pattern breadth
The solver-backed demo pattern is confirmed in TSP and ShowcaseAgent routing. More domains would strengthen the claim.