research-person
Skill6 starsChanged 4 months agoMITView on GitHub
Research a person from public sources into a confidence-marked Obsidian People dossier.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- spirosbax/insaighttheir own version, first seen Oct 13, 2026
- noticedso/skillstheir own version, first seen Aug 24, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- spirosbax/insaight★ 32same name
- noticedso/skills★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository27 other files it gives its agents.
- agent-skills-agentsAGENTS.md
- agent-skills-claudeCLAUDE.md
- agent-log-forensicsSkill
- changelog-writerSkill
- chezmoi-driftSkill
- contacts-unifySkill
- decision-logSkill
- dependency-pinningSkill
- django-smoke-alarmSkill
- empathy-auditSkill
- first-contactSkill
- handle-prSkill
- launch-sequenceSkill
- parallel-isolated-app-testingSkill
- post-mortemSkill
- pre-mortemSkill
- profile-meSkill
- ralph-loopSkill
- release-operatorSkill
- remote-host-verifierSkill
- scope-hammerSkill
- second-opinionsSkill
- status-copy-trust-auditSkill
- support-stormSkill
- trust-auditSkill
- visual-qa-loopSkill
- wide-open-brainstormSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 9 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 22
If no vault exists, write one markdown file in the current directory instead and say so.
- Distinctive: line 46
Don't research minors; don't research someone the user hasn't named.
- Distinctive: line 62
Check the vault first: If a note already exists, that note IS the source of truth for who this is — read it, reuse its identity, and update …
- Distinctive: line 66
If the name is common (e.g.
- Distinctive: line 75
Search/fetch angles (skip what doesn't apply):
- Very specific: line 85
Default to the available web search and page-fetch tools.
- Very specific: line 88
When a thread looks deep or multi-source, you can delegate the heavy searching to parallel subagents (or a deep-research skill, if …
- Distinctive: line 91
Delegation hygiene: When you fan out research to subagents, instruct each to return its findings as its final message — never to write …
- Distinctive: line 170
Keep it honest and skimmable: one idea per line, no padding, no LinkedIn-bio voice.
- Distinctive: line 177
LOW identity match → do NOT write a vault note: Keep the research + ruled-out namesakes in the raw-research dump only, and report back …
- Distinctive: line 185
Use the raw-research folder confirmed in setup.
- Very specific: line 196
Always finish by updating updated: / adding a (researched YYYY-MM-DD) marker so refreshes are visible.
- Very specific: line 212
If the user wants to hand this to a friend, a colleague, or a different agent/tool that can't run this skill, give them the standalone …
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Research Person
Turn a name into a useful, honest People dossier in the user's notes vault — the kind of note that makes the next conversation, intro, or working relationship go better. Built for a personal relationship KB: real friends and colleagues, researched from public sources only, for the user's own private notes.
Paths below are examples from the original setup. On first use, confirm with the user: the vault root
(e.g. ~/cloud/gdrive/obsidian-vault-carlkibler/), the People folders for work vs personal contacts
(e.g. halo/People/ and People/), and the raw-research folder (e.g. ~/dev/agent-notes/<work|me>/people-research/).
If no vault exists, write one markdown file in the current directory instead and say so.
This is the generalized version of a clean OSINT pass: resolve identity → gather public footprint → mark confidence → write it where it belongs without trampling what's already there.
When to use
- "Research
<name>and fill out their People note" - "Build a dossier on these folks:
<list of names>" - Refreshing a stale People note before a meeting, intro, or 1:1
- A new colleague/friend you want context on before working together
When NOT to use
- A public figure for a general report → use a general research workflow (e.g.
deep-research) instead (this skill writes to the People KB) - Building a profile of the user themselves → use
profile-me - You only need one fact (current employer) → just search; don't spin up the whole dossier
Ethics & guardrails (read once, apply always)
- Public sources only. Things the person has published or that are openly indexed. No paywalled background-check services, no scraping behind logins you'd need to deceive to pass, no pretexting.
- For the user's private KB, to support real relationships — not surveillance of strangers. If the named person isn't someone the user plausibly knows or works with, pause and confirm before researching. Don't research minors; don't research someone the user hasn't named.
- Local and private. Notes and raw dumps stay in the user's vault. Don't publish them, post them, or put them in a shared or public repo. Contact details go in only if the user already has them or the person publishes them openly.
- Contain sensitive material. Home address, family members, health, finances, religion, politics,
legal history — even when public — go in a clearly-fenced
## Sensitiveblock, each line marked⚠️, never in the orientation blurb or frontmatter. If it isn't relationship-relevant, omit it. - Identity confidence is a first-class output. A confidently-written note about the wrong person is the main failure mode. Always state how sure you are it's the right individual.
<process>Phase 0 — Disambiguate before researching
- Take the name plus any disambiguators the user gives: employer, city, role, email, a LinkedIn/GitHub/site URL, "the one who…", mutual company.
- Check the vault first. If a note already exists, that note IS the source of truth for who this is —
read it, reuse its identity, and update in place (Phase 4 routing).
- Work:
<vault>/<work-folder>/People/<Full Name>.md(e.g.halo/People/) - Personal:
<vault>/People/<Full Name>.md
- Work:
- If the name is common (e.g. "Mike Smith") and there are no disambiguators and no existing note, ask the user for one or two before burning a research pass. Guessing wrong is worse than asking.
- For a batch, run people in parallel (one research effort each) but keep their notes separate.
Phase 1 — Gather the public footprint
Cast wide, then verify. Aim to answer: who they are, what they do, where they've been, what they've made, how the user knows them, and what's useful for the relationship.
Search/fetch angles (skip what doesn't apply):
- Identity & current role — employer, title, location. Cross-check 2+ sources before stating as fact.
- Professional history — past roles, companies, education. LinkedIn, company About pages, bios.
- Builders' trail — GitHub/GitLab, personal site/blog, Stack Overflow, npm/PyPI, patents.
- Writing & speaking — talks, papers (Scholar/academia.edu), podcasts, posts, interviews.
- Public social presence — X/Twitter, Mastodon, LinkedIn, Instagram/YouTube if public and relevant.
- News & affiliations — press mentions, boards, orgs, communities, open-source projects.
- Shared context — mutual employers/companies already in the vault, mutual people, overlapping history.
Tooling notes:
- Default to the available web search and page-fetch tools. For JS-heavy or fetch-blocked sources (LinkedIn, academia.edu,
Bandcamp/Spotify, some company sites) use a browser-automation tool if one is available and pull
innerText,<meta>tags, and JSON-LD — that's where the verifiable structured data lives. - When a thread looks deep or multi-source, you can delegate the heavy searching to parallel subagents (or a
deep-researchskill, if installed) and curate their findings into the dossier. This skill owns identity-resolution, confidence, routing, and the write; the others are just fan-out engines. - Delegation hygiene. When you fan out research to subagents, instruct each to return its findings as its final message — never to write files. Only the main run writes (the raw dump and the vault note); a subagent that tries to write is sandbox-blocked and wastes a turn. Give each a crisp identity anchor and the exact structured output shape you want back.
- Dump raw findings as you go to the research folder (Phase 4) — URLs, quotes, screenshots. The curated note links back to it; never make the user re-run the search to see your evidence.
Phase 2 — Mark confidence on every non-obvious claim
Per vault convention, tag facts inline:
- VERIFIED — stated in a credible source you actually read (cite it).
- INFERRED — reasoned from evidence, not stated outright.
- UNKNOWN — an open question worth listing (these are valuable; they're the next research).
Plus an explicit identity confidence line near the top: Identity match: HIGH | MEDIUM | LOW with the
basis (e.g. "matched via GitHub→personal site→employer; photo + city consistent"). If MEDIUM/LOW, say what
would raise it.
Phase 3 — Compose the dossier
Match the existing People-note shape. Two flavors:
Work colleague (halo/People/) — frontmatter
---
aliases: [Full Name, nicknames, handle]
tags: [person, <role/team>, confidential]
companies: [[Company A]], [[Company B]]
updated: YYYY-MM-DD
---Personal friend (People/) — frontmatter (no confidential, no work tags)
---
aliases: [Full Name, nicknames]
tags: [person, friend]
updated: YYYY-MM-DD
---Body (both flavors)
# Full Name
> One-paragraph orientation: who they are, how the user knows them, the single most useful thing to know.
**Identity match:** HIGH — basis in one line.
## Identity & contact
- Role / employer / location — **VERIFIED** (source)
- Handles: GitHub `x` · site `url` · LinkedIn `url` · X `@x`
- Email / phone — only if the user already has it or it's openly published
## Background
- Career history, education, notable work — confidence-marked, sourced.
## Public presence & work
- Projects, writing, talks, repos — with links.
## How the user knows them / shared context
- Mutual companies [[Company]] · mutual people [[Person]] · overlapping history.
## Conversation hooks
- Genuine, specific things to connect on (interests, recent work, shared people). Not flattery.
## Sensitive (public but handle with care) ← only if any exists
- ⚠️ item — why it's fenced.
## Open questions
- [ ] Unknown to resolve — **UNKNOWN**
## See also
[[Related person]] · [[Company]]
## Sources
Distilled from: `<raw-research path>` (date). Key public sources: <list>.Keep it honest and skimmable: one idea per line, no padding, no LinkedIn-bio voice.
Phase 4 — Write it where it belongs (auto-write, but never clobber)
Routing:
- Existing note anywhere in the vault → update that file in place.
- Else, work/HaloHealth colleague →
halo/People/<Full Name>.md; personal friend →People/<Full Name>.md. - LOW identity match → do NOT write a vault note. Keep the research + ruled-out namesakes in the raw-research dump only, and report back asking the user for one disambiguator (city, employer, a profile URL, or "the one who…"). A confident note about the wrong person is the exact failure this skill exists to avoid. Promote it to a real People note once confidence reaches MEDIUM+ or the user confirms the identity. (Exception: if a vault note already exists, leave it and just append the open question.)
- Genuinely ambiguous and no signal → ask the user which.
Raw research dump (always): Use the raw-research folder confirmed in setup. Example layout (adjust to the user's):
- Work →
<raw-research>/work/people-research/<name-slug>/. - Personal →
<raw-research>/me/people-research/<name-slug>/. - Save
findings.md(URLs + quotes + confidence) and any screenshots/exports there.
Merge rule — preserve existing content:
- Personal notes are often freeform logs / 1:1 notes. Do not overwrite them. Read the file, then
insert a
## Background (researched <date>)dossier section (and the other dossier sections) above or below the existing log, leaving every dated entry and human-written line intact. - For an existing structured note, enhance in place — update stale facts, add new findings, bump
updated:, append to Open questions. Never drop the user's hand-written lines. - Always finish by updating
updated:/ adding a(researched YYYY-MM-DD)marker so refreshes are visible.
Cross-link both ways: link every Company and Person the note touches; redlink [[Like This]] for
named-but-missing entities (marks a note worth creating). Weave into hubs like [[_People Directory]] when
present.
Phase 5 — Report back
Tell the user, per person: where the note landed (path), identity confidence, the 3–5 most useful new things found, anything fenced as sensitive, and the top open questions. For a batch, a one-line-per-person summary table.
</process>The handoff version
If the user wants to hand this to a friend, a colleague, or a different agent/tool that can't run this skill,
give them the standalone copy-paste prompt in references/handoff-prompt.md.
It carries the same methodology and guardrails without the vault-write step.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

