A suite of agent skills for noticed — a personal networking agent. Each skill wraps the noticed MCP to handle one job in the relationship-management loop: capturing people you meet, debriefing meetings, researching contacts, following up, searching your network, matching identities, and onboarding.
These skills call the noticed MCP. Connect it to your client before you install the skills, or they will not have anything to talk to.
Go to noticed.so/onboarding and pick your client — Claude, Claude Code, Cursor, Gemini CLI / Antigravity, ChatGPT, or anything else that speaks MCP. The MCP server itself lives at noticedso/cli if you want to run it locally.
/plugin marketplace add noticedso/skills
/plugin install noticed-skills@noticed-skills
Requires a Pro, Max, Team, or Enterprise plan with code execution enabled (Settings → Capabilities).
- Open Customize → Plugins and click + → Add marketplace.
- Sync from this GitHub repo:
noticedso/skills. - From the synced marketplace, install noticed-skills. All seven skills land in one shot.
ChatGPT has no GitHub marketplace install, so each skill goes in by hand.
- Open Skills → New skill → Upload from your computer.
- Upload each
skills/<name>/folder from this repo as its own skill (zip the folder first if ChatGPT asks for a single file). Repeat for all seven.
Workspace admins: confirm Enable skills and Enable skill uploading are on under Permissions & roles.
| skill | what it does | writes? |
|---|---|---|
| add-person | Add or update one or more people in noticed from a name, URL, handle, photo, or freeform dump — dedupe-first, logs the meeting on the day it happened, saves context as a memory. The core capture primitive. | yes |
| event-debrief | Process a meeting/event dump into people, notes, follow-ups, ideas, and references — all written in one pass. | yes |
| research-person | Deep-dive one person before a meeting; render a dossier from noticed + the web; offer to save it back. | only on save |
| follow-up | Draft a follow-up message to someone you just met, in your voice, with something actionable in it. | logs the touchpoint only after you confirm it went out |
| search-network | Natural-language search across your network; returns a tight table; drills into anyone. | read-only |
| onboard | Onboards you to noticed: a short questionnaire that captures your identity, goals, and current tools, saved in one save_onboarding call. |
yes (to your own record) |
| match-identities | Triage the cross-source identity matches noticed proposes — confirm real matches, clear false positives, merge duplicates — and fill profile gaps by searching your network first. | yes |
The capture loop runs across a real-world interaction:
add-person → event-debrief
(during) (after)
research-person and follow-up support any contact at any time. search-network is the read path over everything you've captured. onboard is first-run setup.
All skills follow the same rules, so they read as one system:
- Identity resolution is owned by
add-person(own-network search → web enrichment → ask). The other skills follow that flow rather than re-implementing it. - Provenance, two surfaces. In chat, web-found facts are attributed softly ("his linkedin says…") so research never reads as something you said. In the stored note, every line carries a system-only tag (
[from user]/[research, unverified]) that is never shown in chat. - Readback after every write. noticed writes are silent, so each skill recaps what landed — in prose, not a database-style
tags:row. - No public scope in the capture path. Capture answers who is this person via own-network + web. Warm-intro / reachability discovery (
scope: "public") lives only insearch-network, where it falls back to own-scope silently if the backend errors. - No Gmail dependency. Drafts are copy/paste (subject included for email); nothing is sent automatically.
See TESTING.md for suggested prompts to exercise each skill and a checklist of behaviors to confirm.
At the end of a session that used a skill, paste the prompt in skill-improvement.md to have the agent review how the skill actually behaved and suggest edits — biased toward making it simpler first, then more accurate. Works in any client and for any of the seven skills.
Updates are manual for now. Reinstall to get the latest.
Releases are versioned automatically: any change under skills/ on main triggers CI
(.github/workflows/version-bump.yml), which bumps
the version in marketplace.json (semver, derived from the conventional-commit messages) and
publishes a GitHub Release.
To pull a new version, refresh the marketplace — reinstalling the plugin alone reuses a cached copy and will not see new commits:
/plugin marketplace update noticed-skills
Then update the plugin (the Update button activates once the refreshed marketplace reports a higher version). On claude.ai / ChatGPT, re-sync the marketplace from the repo rather than reinstalling the plugin.