EllypsisAI

eval-rerun

Re-runs one case from an existing mcp-eval test plan with a fresh sub-agent, then compares the two runs. Use only when an eval has already been run against an MCP server and the user wants a single case repeated — checking whether a result is stable or flaky, asking whether the eval is deterministic, or verifying that a fix to a tool description or schema actually worked. Not for starting a new evaluation and not for re-rendering the dashboard.

eval

Evaluates whether an MCP server is production-ready with the LLM that will actually drive it — tool discoverability, argument fidelity, payload economics, reliability, and grounding — then writes an HTML dashboard with a ship / don't-ship verdict. Use this whenever the user wants to evaluate, test, benchmark, audit, vet, or review an MCP server or its tools; asks whether an MCP is good, production-ready, or worth shipping; wonders why Claude isn't finding or using a connected server's tools; asks what a server's responses cost in tokens or context; or is comparing two MCP servers. Use it even when the request sounds casual ("can you take a quick look at this MCP", "does this server actually work with Claude") — the shortcut version of this work produces a confident score of nothing.

Data & Documents Listed

eval-report

Re-renders the mcp-eval HTML dashboard from artifacts already on disk in an eval workspace, running no tests and grading nothing. Use only when an eval workspace already exists and the user wants the report rebuilt, reopened, or regenerated — after hand-editing a verdict, after re-running a case, or to look again at a report from an earlier session. Not for running an evaluation and not for repeating a test case.

contacts

This skill finds decision-makers and contacts at target companies. Use when the user asks to find people, look up who works somewhere, or identify who to reach out to. Trigger on: 'find contacts at X', 'who should we reach out to at X', 'find the right person', 'get me contacts', 'who's the fleet manager at X', 'look up people at X', 'find their procurement team', 'who can we talk to there'. Also use after company research when the natural next step is identifying people to contact. Do NOT use for writing emails (that's outreach) or logging contacts to CRM (that's the crm skill).

crm

This skill handles all CRM operations — creating records, logging data, and helping salespeople execute their task queues. Use when the user wants to persist anything to the CRM or work through their sales tasks. Trigger on: 'log it', 'save to CRM', 'create the org', 'add them to the CRM', 'create the opportunity', 'push to CRM', 'log the contact', 'save this', 'save the draft', 'save that draft for later'. Also use when the user brings up their task list and wants help working through follow-ups, or says 'mark it done', 'what tasks do I have', 'help me with my follow-ups'. This is the only skill that writes to the CRM — research, contacts, and outreach produce the content, this skill persists it. Do NOT use for researching companies (that's research), finding people (that's contacts), or drafting emails (that's outreach).

outreach

This skill drafts personalized sales outreach — intro emails, follow-ups, LinkedIn messages, and full email sequences. Use when the user asks to write, draft, or compose any sales communication. Trigger on: 'draft an email to X', 'write outreach for X', 'compose a follow-up', 'write a LinkedIn message', 'draft the sequence', 'write an intro email', 'help me write to this person', 'prepare the emails', 'what should I say to X'. Also use when research and contacts are ready and the user wants to act on them by writing messages. Do NOT use for finding companies (that's research), finding people (that's contacts), or saving drafts to CRM (that's the crm skill — though outreach produces the draft content that crm stores).

research

This skill is the research method — deep research on anything without flooding the main context. Use for any research request: 'research X', 'look into X', 'what do we know about X', 'find everything about', 'deep dive on', 'find me all', 'is X a good fit', competitive analysis, market scans, or when the user shares a company name or URL and expects intelligence gathering. Also use when a task needs research as context. Do NOT use for finding people (that's the contacts skill), writing messages (that's outreach), or logging results (that's the crm skill).