| name | resonite-live-send-debug |
| description | Run and debug PLATEAU-ResoniteLink live-send reproductions against a real ResoniteLink session. Use when the user wants actual machine-level validation instead of simulated tests, including listener discovery, targeted slot removal, log capture, and inspection of the resulting Resonite world state. |
Resonite Live Send Debug
Use this skill only for real ResoniteLink runs. Prefer local tests first, then switch to this skill when the question depends on a live session, a destructive cleanup cycle, or the resulting Resonite world state.
This file is the Coding Agent entrypoint for the live-send workflow in this repository and the authoritative live-send workflow reference for the public operator surface. Keep detailed operational guidance in references/workflow.md and keep this file focused on trigger, guardrail, and output contracts.
When To Use
- Real live-send reproduction against an actual ResoniteLink listener.
- Validation that requires observing logs, process state, or the resulting live world.
- Targeted slot removal, slot dumps, or headless-session bring-up as part of the verification loop.
When Not To Use
- Code-only review, static log reading, or documentation work.
- Local/unit/integration tests that already prove the issue without a live session.
- Any task where destructive cleanup of the current dataset root is not acceptable.
Guardrails
- Treat slot removal as destructive. It can remove live content from the current world.
- Do not ask the user to run the live send if you can run it directly.
- Do not compare runs until targeted removal has been verified and a post-removal pre-send root dump confirms the base world state.
- Do not run cleanup, root-dump, and live-send steps for the same world in parallel. Removal verification, pre-send dump capture, and each live-send run must stay serialized.
- Do not encode dataset-root, shared-assets, or common-material naming semantics into the tool. Use
dump-slot and remove-slot as thin primitives only.
- Keep the final successful
DatasetRoot in place unless cleanup is explicitly requested.
- Treat interrupted or partial runs as provisional unless cleanup and post-run state were both verified.
- Use direct
dotnet commands as the operator surface. Do not recreate thin wrapper scripts or a project-based session tool.
start-headless is part of the direct tool surface. If --headless-path is present, resolve the launcher from that path and a few standard nearby candidates only. If --headless-path is omitted, try only the standard Windows Steam install root, then let the actual launcher execution decide whether the environment supports it.- Do not substitute unrelated machine-local copies when the configured headless path is wrong. If the provided path does not contain a launcher, stop and point to the standard installed-path guide in
references/workflow.md.
- This skill does not bridge environments. If ResoniteLink uses
localhost, run sender, listener, and headless in the same environment and document that assumption in the run notes.
Guide Surface
Use the guide for:
- recommended datasets and fixture values
- fixed run worksheets and comparison checklists
- component discovery and BoxCollider inspection procedures
- version-scoped readback limits and reference artifacts
- direct command examples for the CLI and session tool script
Operator Surface
Use only these operator-facing direct commands:
dotnet run --project src/PlateauResoniteLink.Cli/PlateauResoniteLink.Cli.csproj -- import ...dotnet .agents/skills/resonite-live-send-debug/tools/session-tool.cs -- discover-session ...dotnet .agents/skills/resonite-live-send-debug/tools/session-tool.cs -- dump-slot ...dotnet .agents/skills/resonite-live-send-debug/tools/session-tool.cs -- remove-slot ...dotnet .agents/skills/resonite-live-send-debug/tools/session-tool.cs -- start-headless ...dotnet .agents/skills/resonite-live-send-debug/tools/session-tool.cs -- stop-headless ...
In sandboxed environments, these direct commands can still require restore/build escalation. If dotnet restore, dotnet run, or dotnet <script>.cs fails on .NET first-use or permission setup, rerun the same direct command with the required sandbox escalation instead of replacing it with an ad hoc workflow.
Required Outputs
Summarize each live run with:
- listener endpoint
- slot-removal verification result
- post-removal pre-send dump result
- process status and exit code
- exact mode and mesh code
- last timestamped
import line
- last timestamped
live line
- whether
stderr was empty
- world snapshot summary
- root dump paths
- observation timestamps
- whether the conclusion is valid or contaminated
