Business & Operations

RAG-backed. Take a DipTrace board through requirements, source evidence, schematic, PCB, production handoff, bring-up, and revisions using existing project skills and explicit build gates. Use for a full design cycle, a new board specification, or resuming a PCB build; «полный цикл платы», «ТЗ на плату», «продолжи разработку». Use when the user says “Take this board through the complete design and production cycle.”

RAG-backed. Collect reviewable DipTrace evidence with native headless PCB acceptance when supported, open/save/close for all four editors and supported native capture; for unsupported actions guide an operator through quarantined source/open-save/re-export capture, dry-run ingest, MCP validation, explicit confirmation, and metadata recording. Use when the user says “Guide an operator through a reviewable DipTrace round-trip capture.”

RAG-backed. Plan and optionally commit one bounded critical-net or differential-pair route in DipTrace with explicit rules and post-checks. Use when the user says “Route this explicitly named critical PCB net.”

Plan and place physical PCB test points (standalone probe pads) on a DipTrace board through the DipTrace MCP server — coverage audit, free-grid candidate selection, guarded dry-run placement, stub routing, and post-checks. Use when the user asks to add or place testpoints on a PCB, improve probe access or testability, plan fixture/bed-of-nails points, or audit PCB test-point coverage.

Improve Coval trace quality after basic ingestion works. Use when traces are sparse, missing useful STT/LLM/TTS/tool spans, missing attributes needed for Coval built-in metrics, or when a customer wants maximum debugging and observability value from agent traces.

Delegate a read-only Coval evaluation, simulation, monitoring, product-how-to, or troubleshooting subtask to Sofia through the official Coval MCP consult_sofia tool. Use when the user needs Coval-specific FDE judgment, analysis of organization data, metric or test-set recommendations, or a diagnosis grounded in Coval runs and conversations.

Launch a specific Coval evaluation with explicit cases and execution bounds. Use when resources are already selected; for result analysis use quick-eval.

Select and configure a Coval metric for a concrete product criterion, choosing transcript, audio, trace or deterministic evidence and testing the result. Use for metric creation or scoring design, not for final human calibration.

Analyze development-set disagreements between human annotations and one Coval text LLM judge, then propose a focused prompt revision. Use coval-calibrate-metric for independent trust measurements.

Compare before/after Coval runs using matched cases, stable metric versions and inspected conversation evidence. Use to assess a regression or proposed improvement; does not automatically launch or tune anything.

Validate one Coval judge against independent human labels, with grouped development/test separation, class-specific errors and uncertainty. Use for trust claims or calibration; without real labels, prepare review rather than inventing ground truth.

Work Mesh Live — automatic presence via hq mesh daemon; context reconcile/organize; manual session task-status, blocked, and note only.

Take an idea all the way to a finished deliverable in one command — capture it, research and compare approaches, generate the PRD, then execute every story to the actual work product (code, artwork, documents), continuously and without pausing. Recommended answers are taken automatically and recorded in the decision ledger; --gated pauses at human decisions instead; --plan-only stops at the execution-ready PRD. Use when the user says "/orchestrate", "take this idea to done", "make this end to...

Scaffold a new HQ company AND optionally take it all the way to operational — business-discovery interview, seeded knowledge/workers/skills/projects, brand design packs (generated from website/PDF/Drive and bound to deploys via policy), connected integrations, org groups + ACL rules, teammate invites, and optional cloud agents.

Provision a fleet agent end-to-end — identity, membership, vault join, secrets, file access, MCP/runtime bootstrap, mission brief, and a verified capability probe. Use when standing up a new HQ agent (Slack bot, reporting agent, ops agent) or when an existing agent reports it is blocked on access.

Inspect or change HQ vault ACLs, grants, and shared prefixes.

Turn a resolved ticket into a help-center article, draft a known-issue status page when something breaks, send personalized 'you asked, we shipped' notes to the customers who requested a feature, or flag stale articles that need a refresh after a product change. Pick the flavor and get a publish-ready draft grounded in real conversations and your voice.

Update the rules that decide how incoming tickets get sorted and where they go. Change what counts as a bug versus a feature request, point bugs at a new tracker, update who approves refunds, add a new category, or adjust VIP routing. Every triage after the update automatically follows the new rules.

Hand me a new ticket and I sort it for you. I read the message, figure out what it is (bug, how-to, billing, feature request), check if the customer is a VIP, assign a priority based on your routing rules, and slot it into your queue so you know exactly what needs attention and how fast.

Every time you tell a customer you'll do something by a certain date, I write it down with a deadline so it doesn't slip. I pull the promise straight from your reply, parse the due date, and link it to the thread. These show up in your morning brief automatically, so nothing falls through the cracks.

Tell me the basics about your product, your customers, and how you handle support so I can give you better help. I ask a few quick questions about your product, response-time targets, VIP list, routing rules, and known gotchas. You only need to do this once, and I keep it updated as things change.

Get a structured readout of how support is going. A weekly rollup that covers every domain with volume, top themes, overdue promises, and churn flags opened. A help-center digest that surfaces ticket themes, request velocity, and the single most useful article to write next. Or a per-account review that maps wins, shipped requests, open friction, and next moves so you walk into the call prepared.

Give me a customer name and pick what you need: a dossier with their plan, history, and open items; a full timeline of every interaction; a health score (green/yellow/red) with the three signals driving it; or a churn-risk check that tells you if they're slipping and what to do about it. One customer, four angles.

Spot something in a ticket that's bigger than the ticket itself and I file it properly. A bug report gets logged with repro steps and severity so engineering can act on it. A feature request gets attributed to the customer who asked. And if I see the same question come up three or more times without a help article, I flag the docs gap and offer to write one.