Every answer, with its sources.
A skill for research you can check. Your agent answers from the sources you hand it, or from the web, with a citation on every claim and a plain note on what it could not cover. If you already keep a private place for evidence, it can save what it used there too. Hunter ran it live in the Campaign Pipeline drop.
Most agent research reads well and is hard to check. Which page said that, and when? r2c makes your agent keep track: where each fact came from, when it was pulled, the exact spot in the source, and how much the sources really covered. It leads with the answer, shows what supports it, and says what it could not find. Saving is a separate step, and it is optional. With no saving set up, r2c researches and answers, and that is the whole skill. With a private evidence folder, or a Cairns vault you already run, it can keep the sources it used so the next question starts ahead.
/r2c, a research skill that cites every claim and names its gaps. An optional save route that locks the evidence it used and hands it to one writer. A guide for saving into a Cairns vault, read only when saving is allowed.
You can check any claim in a minute. Five reposts of one article count as one source, not five confirmations. A missing transcript stays missing instead of being guessed at.
It works out which mode it is in from what you have allowed: research only (the default), save through the private route you already approved, or no saving because you said so. No saving always wins.
Give it the question, and the sources if you have them. Decide where evidence may be kept, or that it may not. Read the citations on anything you are about to act on.
Three modes, and the one it starts in.
- Research only. The default. An answer with citations and its limits. No Cairns, no paid tools, no setup.
- Save through your approved route. Only when you have already said where evidence may go, a private folder outside anything shared or synced or your own Cairns, and which steps it may do. Asking for r2c is not that yes.
- No saving, read only, or off the record. Overrides everything. No files, no checkpoints, no background writer.
The skill was written to feed a Cairns vault, the layered second brain from the YouTube Cairns blueprint: the untouched originals at the bottom, one reviewed card per source above them, an index on top. You do not need one. Without it, r2c is a research skill that shows its sources.
What happens when saving is on?
Three steps, and the skill does not call anything saved until the last one passes.
- Lock the evidence. What it used is frozen as one packet: the original, its date, its link, the exact spot, how sensitive it is. Originals stay untouched; cleaned text is kept as a separate copy that says where it came from.
- One writer. A single background worker does all the writing to your destination, so two agents never edit the same index at once. Later packets go to that same writer.
- Check what came back. Each layer has to check out on its own: the original is stored (L3), its card passed review (L2), and the index link reads back (L1). An upload, or a green finish, proves none of those.
The answer with its citations. Then, if saving was on, a plain list of what is saved, what is pending, and what is held for review. A card that failed review stays held; it is never quietly marked ready.
What you can ask your agent, once it is on
Start with the first one. Copy, paste, read what comes back.
Use r2c to compare these supplied sources. Answer with citations. No saving or subagents.
Use r2c to research this question: [your question]. Answer with citations, and tell me what you could not cover and how fresh the sources are. Do not save anything.
Save the selected public evidence in the private local folder I specify: [folder]. Verify the files and hashes; do not claim Cairns completion.
Use r2c to research this question. Reuse my existing approved private saving policy and permitted stages, with one background writer if delegation is authorized. Report verified saved and pending stages separately. If that policy is absent, research only; do not configure services.
From that r2c answer, list the claims that rest on a single source, and the sources that are the same story copied.
Words you now own
- Citation. Where a claim came from: the source, its link, the date, and the exact spot inside it.
- Source family. Copies of one original, like clips of the same video or an article reposted on three sites. They count as one source.
- Coverage. How much of the question the sources actually answered. The skill says it out loud rather than implying it found everything.
- Evidence packet. The locked bundle of sources an answer used, with dates and fingerprints, so a save can be checked against it.
- One writer. The single worker allowed to write to your evidence store, so saves never collide.
- L1, L2, L3. The Cairns layers: L3 the stored originals, L2 the reviewed card per source, L1 the index that points at them.
Before you rely on it
Three habits, because a citation only tells you where to look.
- It checks that a claim matches its source, not that the source is right. A cited claim from a bad page is still a bad claim.
- Pages change. Each source carries the date it was pulled, so you know how fresh the answer is. Re-run it before you lean on an old one.
- You choose what goes into a search. The skill keeps private evidence out of public searches; you still decide what to paste into the chat.
Research to Cairns can use Exa, Firecrawl or Scrape Creators with your own account. If one needs a key, you enter it yourself through that service's own secure sign-in or Infisical, never in the chat, a file, a command or a log. See Keys, the First Time.
lab/ or the skill's own folder.Evidence can hold private material, so it goes in a private folder you name. The Lab's own folders are replaced on every update, so anything saved there would be lost.
Fair questions
No. Research only needs the sources you hand it or the web search your app already has. Cairns, paid services and new accounts are all optional, and the skill never sets them up for you.
Not on its own. It uses the tools you already have, inside the limits you already set; a new subscription or a bigger budget needs your yes. For scale, the outside research in the Campaign Pipeline session cost $0.03 in Exa plus five Firecrawl credits.
A skill is a folder with a SKILL.md, and both apps read that format. In Codex you call it with a dollar sign instead of a slash.
In Campaign Pipeline, drop one. Scratch used r2c for the outside sources behind Sara's campaign, and the research was saved for the next chat.