Agent Runtime Tools
Use these tools to separate source evidence, automated contract evidence, and live-client observations. A source read or passing test is not a substitute for observing a revision-sensitive flow in RuneLite.
Current Offline Tools
| Tool | Use | Notes |
|---|---|---|
| CodeGraph | Find source symbols, callers, and file structure before editing | Treat index output as navigation, then verify changed files after edits |
rg | Search docs and source text | Use targeted patterns; avoid broad scans after the symbol is known |
| Gradle focused suites | Compile and test narrow areas | Start focused, then broaden when shared code changes |
git diff | Review exact edits before final notes | Check docs and prompt changes for stale paths |
docs-site/sidebars.js | Confirm published navigation | Canonical pages belong in the sidebar or need an intentional inbound link |
Current Runtime Inspection Surfaces
| Surface | Package | Use |
|---|---|---|
SuiteRuntimeStatus | InteractionApi.debug | Snapshot revision, bootstrap, pacer, walker, and health status |
WidgetExplorer | InteractionApi.debug | Inspect widgets by visibility, text, and action |
PacketTrace | InteractionApi.debug | Capture bounded packet trace events when enabled |
RevisionHealthCheck | InteractionApi.debug | Probe revision-sensitive packet reflection surfaces |
| Plugin overlays and panels | plugin packages and com.n3plugins.ui | Read live plugin status and user-facing state |
| RuneLite logs | local RuneLite runtime | Correlate plugin errors with live client state |
| API Tester | com.n3plugins.apitester | Run curated, user-visible SDK/API probes inside RuneLite |
| Agent Server | com.n3plugins.agentserver | Expose allowlisted runtime reads and diagnostics to a local client |
Use these as evidence helpers. They do not replace source review or live observation.
Offline Agent Loop Available Now
- Read
AGENTS.md, SOT, and relevant docs. - Map the requested workflow to existing SDK/actions.
- Inspect live source with CodeGraph or
rg. - Patch only scoped files.
- Run focused compile/tests.
- Inspect
git diff. - Update docs, SOT, and sidebar/cross-links when public behavior changes.
- Report commands, results, and live verification gaps.
Runtime Evidence Guardrails
- Local-only by default. No network listener beyond localhost.
- No account credentials, session tokens, or personal paths in artifacts.
- No hidden screenshots or logs. State output paths before writing.
- Prefer read-only probes. Do not use a diagnostics endpoint for risky, destructive, costly, or account-specific gameplay actions.
- No generated artifacts committed unless the user requests them.
- Record client revision, setup, action or probe, expected result, observed result, and an evidence reference for every required live scenario.
Runtime Tool Review Checklist
- Does the tool read or write files? State paths before running it.
- Does it rely on the live client? Record client revision and account state.
- Does it expose screenshots or logs? Redact sensitive content.
- Does it only inspect state? Keep diagnostics separate from action dispatch.
- Does it prove live behavior? If not, mark live verification pending.