skills

Onboard a new user to noticed and gather light user research at the same time. A short chip-driven questionnaire (about 7 questions, 2-3 minutes) that captures basic identity (role, location, what they're building) and research about how they manage relationships today + what they want from noticed — then stores everything in the noticed system with a single `save_onboarding` call. Pre-fills from `my_profile` and `get_person('me')` to skip questions the agent already knows. Trigger on "onboard me", "set me up", "interview me", "noticed onboarding", "what does noticed know about me", or any first-run intent.

Add or update a person in noticed, dedupe-first. Takes whatever the user provides (LinkedIn URL, name, handle, freeform context, batch from an event), resolves against the existing network — updating in place when found, creating only when not — logs a `met` interaction with the right date by default, and saves freeform context as a durable memory. Trigger on "add X", "save X", "put X in noticed", "remember X", a pasted LinkedIn/GitHub URL with context, or a batch from tonight's event.

Process a meeting or event dump into structured pieces in noticed: people, meeting notes, follow-ups, ideas, references. Use whenever the user pastes a debrief from any meeting (1:1 call, walk, dinner, group event). Trigger on "debrief", "process this voice note", "here are my notes from [event]", "I just had a [meeting / walk / dinner] with [person]", or any post-meeting dump. Parses the dump, resolves each person against the user's network, builds a unified preview, and writes everything in one fan-out on the user's go-ahead.

Draft a follow-up message to someone the user recently met IRL. Figures out new vs existing in their network, picks the right channel (email, linkedin, whatsapp, telegram, sms), suggests an angle from context, drafts in the user's voice with something actionable in it. Returns the text to copy/paste (subject too, for email) plus a link to open the chat. Logs the touchpoint only after the user confirms it went out. Use whenever the user says "follow up with X", "draft a message to X", "thank X for the meeting", "send X our deck", or similar.

Research one person before a meeting or call. Pull together what's already in noticed and what the web says, then render a concise, scannable dossier in chat (overview, interaction history, company context, conversation starters, sources), and proactively offer to save the enriched context back to noticed. Use whenever the user says "research X", "prep me for my meeting with X", "what do we know about X", "deep dive on X", "pre-meeting brief on X", or "tell me about X". One person at a time. Works whether the person is in noticed or not.

Triage the cross-source identity matches noticed proposes for the user's network: confirm the ones that are the same person, clear the false positives, and merge obvious duplicates. Pulls the review queue with `list_identity_matches` (status: 'pending') and acts on each row with `accept_identity_match` (merge), `mark_different_people` (durable "not the same person"), or `dismiss_identity_match` (soft "not now"). Use whenever the user says "review my matches", "clean up duplicates", "are these the same person?", "any duplicate contacts?", "approve/reject the identity suggestions", or asks to tidy who's-who. Also fills profile GAPS, email-only contacts with no LinkedIn/GitHub, via `list_profile_gaps` + `add_profile_to_person` ("add a profile for X", "here's the LinkedIn for that contact"). Writes to the network (merges + disconnects), so it confirms before merging anyone who isn't an obvious duplicate.

Find people in the user's noticed network through natural-language queries. Wraps `search_people` and returns a compact markdown table of 5-10 results. Use whenever the user is looking for someone they might know: "do I know anyone at X?", "find me someone who does Y", "who in my network is in Z?", "any AI engineers?", "search my network for...", "is there a warm intro to X?", "who do I know in [city]?", "who's a founder I'm close to?". Also handles "how many..." network-shape questions via `network_summary`, drills into a single result via `get_person`, and escalates to `scope: 'public'` when the user asks for warm-intro paths or people outside their network. Read-only — no writes. Default scope is the user's own network.