Safely commit and push intended changes from a mixed or dirty worktree. Use when the user asks to push updates, commit current work, publish a fix, or handle git status with unrelated local changes, nested repositories, ignored generated files, or possible index.lock issues.
Execute an approved implementation plan end to end. Use when the user says Implement the plan, points to an execute-and-confirm workflow, asks to complete a named plan, or expects code changes plus verification, changelog/troubleshooting/docs updates, and accurate plan archival or status updates.
Triage saved runtime artifacts before editing code. Use when the user asks to inspect project/execution, project/errors, console.md, terminal.md, output.md, saved logs, warning output, large log-like files, or asks whether a warning/error needs a fix.
Debug and verify exported/rendered output parity across app view, HTML, Markdown, PDF, ZIP, and preview surfaces. Use when thumbnails, image links, rendered PDF, full export ZIPs, markdown previews, blog exports, or generated document styling differ from the app view or fail during export.
Launch or restart a local chat bot and verify real webhook health, auth, tunnel routing, and inbound/outbound delivery. Use when the user says launch my chatbot, restart after changes, verify health, run a real WhatsApp/Kapso/ngrok test, check what model/auth path is in use, or ensure home-env credentials are loaded.
Orchestrate multiple independent agents or model passes to produce a stronger implementation plan. Use when the user asks to launch multiple agents/models, use opencode in single-line mode, run independent parallel research/proposals, add an adversarial critique stage, or synthesize a coding-agent-ready handoff plan.
Verify current AI provider model names, capabilities, API contracts, and routing claims against live provider catalogs, official docs, or the current adapter code. Use when the user asks for a complete accurate provider/model list, whether a provider supports a capability, whether a model name is correct, or whether docs/plans/constants match current provider reality.
Trace a user-facing AI generation setting or provider behavior end to end through UI, state, request schemas, API routes, factories, adapters, and tests. Use when the user says a control must work for all providers, asks to pass parameters to API calls, reports provider selection being ignored, or needs direct vs research generation routes compared.