Fix the bug.
See the evidence.

Local provider implementation; model/backend dependent.

Native arbitrary local-model/offline support unknown from inspected docs; compatible gateways documented.

--oss supports Ollama and LM Studio.

Local llama.cpp router and other supported local endpoints.

Explicit lan/ provider, endpoint discovery and hostname handling. Campaign LAN route remains NOT_RUN.

Gateway base-URL overrides documented; arbitrary LAN inference support unknown.

Custom provider base_url allows compatible network endpoints; LAN discovery/catalog behavior unknown.

Custom provider endpoints; automatic LAN discovery is not established.

Possible with ready local models and local tools; service-dependent features cannot be assumed offline.

Fully offline coding unknown from inspected official docs.

Local OSS inference documented; fully disconnected end-to-end workflow not tested here.

Local inference plus offline network controls; disconnected coding not tested here.

Project trust; per-tool approval requires an extension or external controls.

Seatbelt exists for portable workspace Bash and specific proof oracles. General permission gates do not establish universal process isolation.

Shell sandbox on macOS/Linux/WSL2; off by default; file tools, MCP and hooks outside it. Native Windows commands unsandboxed.

External container or sandbox; Pi tools otherwise inherit process permissions.

Plan workflow available through extensions or packages; no built-in plan mode.

Can add search/fetch through MCP or extensions; dedicated built-in tools not established.

Subagent workflows through extensions or tmux; not built into the core.

/export <path.json|path.md> writes the conversation locally inside the current workspace. Existing files are protected unless --overwrite is explicit. Source-inspected; broader consumer checks remain partial.

/export [filename] exports the current conversation as plain text; a supplied filename writes directly to a file.

A built-in full-conversation local export command is not established by the inspected docs. /copy copies a reply; JSON run events are a different scope.

/export [path] writes the current session to HTML or JSONL. Separate sharing commands can upload a session; local export is distinct.

/rewind alias /undo; tracked-change and Git fallback implementation. Not a universal rollback guarantee.

Checkpoint rewind for tracked direct edits; Bash/external changes and most subagent edits excluded.

Conversation edit/fork documented; equivalent automatic file checkpoint rewind unknown. Docs recommend Git checkpoints. ,

Conversation tree recovery; automatic file undo not established.

/cost, /usage, usage files, provider/Gateway receipts; campaign exposed incomplete coverage.

Token ceilings, verification time/credit controls and bounded retries implemented; scope-specific.

--max-budget-usd, --max-turns; estimated-spend cap can overshoot.

Equivalent universal per-run dollar cap unknown from inspected docs.

A universal per-run dollar budget cap is not established by the inspected docs.

Opt-in check/repair workflow, structured proof result and evidence/usage disclosure. C-S6 remains PARTIAL, not PASS.

No verified equivalent to Verdict's specific receipt/terminal contract; /verify provides model-driven app verification.

No verified equivalent to Verdict's specific receipt/terminal contract; coding/tests/review documented.

An equivalent proof result and terminal evidence contract is not established.

/opinion: multi-model advisory synthesis, presets/models/Decision adjudication. Verdict Server dependent.

Experimental /advisor consults a second model; provider/model eligibility restrictions apply.

Multi-agent/model roles allow separate reviews; dedicated fixed Opinion-style product unknown.

Custom extensions can orchestrate additional models; dedicated consultation UI not established.

/fusion: tool-free one-shot fan-out/merge; feature-gated; checks selected architecture; cloud-direct route unsupported.

Arbitrary multi-provider fan-out/merge equivalent unknown; advisor/subagents offer related workflows. ,

Dedicated Fusion equivalent unknown; subagents offer related orchestration.

Custom extensions can compose model calls; dedicated fan-out/merge product not established.

Small scoring-model features for logs, completion claims, stalls, risk, incidents and sizing; admission/readiness gates. Advice does not certify work.

Dedicated equivalent portfolio unknown; advisor, hooks and review offer related mechanisms. ,

Dedicated equivalent portfolio unknown; subagents/hooks/review offer related mechanisms. ,

Classifier primitives are documented; an equivalent built-in Decision portfolio is not established.

/ide VS Code bridge for context, diffs and session sync. Implemented, not exercised here.

SDK/RPC embedding documented; official companion IDE integration not established.

Cloud/provider routing exists. Full CLI-to-cloud coding-task dispatch lifecycle not established by this inspection.

CLI cloud dispatch plus hosted/self-hosted cloud sessions.

codex cloud submit/list and codex apply.

Dedicated CLI-to-cloud coding-task dispatch lifecycle not established.

/remote-control requires Verdict Server, bridge daemon and cloud redemption. Installed cross-device flow not tested here.

CLI Remote Control continues local sessions through browser/mobile companions.

CLI remote-control commands documented; mobile companion setup starts in desktop app, with rollout/account gates. ,

Terminal/tmux and RPC integrations; equivalent mobile session handoff not established.

Script/CI automation documented; desktop scheduled tasks separate surface. ,

Scripts or extensions can schedule work; a built-in /loop is not established.