git:20261019.7e4f2a3 to git:20261019.376a0bd

9 added, 7 removed. Audit A to A.

---

name: rem-enrich

description: Find the leads a page has not followed — links, attachments, names, older threads — follow them, and add what they establish.

---

# Enrich one page

Input: a written page, its reading ledger and search coverage.

Output: the page with new cited facts, and a lead list in your final reply.

Follow the page kind's `rem-page-*` rules and `rem-investigate`'s quote and

evidence rules unchanged.

Enrich is discovery. The page was written from what init happened to read;

your job is to find **what else exists** and where it points. Finding a

direction to investigate is worth more than polishing a sentence.

## Leads from the last round

The prior-runs item holds the last enrich's open leads and the last muse's

open questions. Follow those first. Mark each followed, open or blocked again;

do not repeat a lead already followed.

## A project's history

- On a project page, the ledger's `github:` items (its own repository and the

- search of the owner's other repositories by its name) and the git log are how

- it was built. A page with no `Key decisions` older than its latest weeks has

- not read them: read the issues and pull requests that name the project, oldest

- first, and add the turning points (when it started, what was merged, what was

- decided and why) as dated, cited `Key decisions` lines. The latest message is

- rarely the lead to follow first on a thin project page.

+ The timeline item is the project's whole history, oldest first, month by month.

+ Walk it from the first month and read the originals at its `github:` ids;

+ look up commits for their full `git:` ids. The timeline itself is a reading guide.

+ Add the turning points — the start, what was merged, what was decided and why —

+ as dated, cited `Key decisions` lines spread across the whole history, not only

+ the latest weeks. For a monorepo sub-project, its history is the pull requests

+ in the timeline whose titles name it. Follow those through the older months

+ before following the latest session.

## 1. Find the leads

Read every cited original whole, then list what it points to that the page has

not followed. A lead is anything that could hold more:

- **A link**: a website, LinkedIn or GitHub profile, a shared document, a

repository, a calendar or booking page, a product or company site. Open it.

- **An attachment** not yet read: a contract, invoice, CV, deck, screenshot.

- **A name**: a person, company, project, product or place mentioned but

without a page or not linked from this one.

- **A thread** whose start or end the page does not have: `Re:` with no

original, a request with no reply, a reply quoting earlier mail.

- **An older period**: first contact earlier than the page's First contact;

mail before the window init read.

- **Another channel**: the same person in sessions, GitHub issues, WhatsApp,

another mailbox or address. The subject's agent mailboxes and aliases

(`<name>@mail.openonion.ai`, `0x…@mail.openonion.ai`, a relay or handoff

address) are their channels too: search mail from and to each one as you

would their own address.

- **A number or date** that implies another record: an invoice number, an

order, a meeting, a deadline someone must have acted on.

- **The reading ledger itself**: every kind of source it lists that the page

cites none of (sessions, GitHub items, attachments, another address) is a

lead; open at least the ones about this subject in each kind.

## 2. Follow them

Most promising first: what could change the lead, Open threads or Facts.

- Open links and read the page; a URL you read is a source (`[n] <url> —

<date read>`). Note what it says about the subject, never what it implies.

A quote from a web page is one run of its text as written; never join two

fields or labels into one quote.

- Search the archives (`rg -il`, `co rem search`) for every name, number and

subject line; read matches whole.

- Read unread attachments; cite their ids.

- Stop a lead when it is empty or turns out to be about someone else; record it

as followed so the next run does not repeat it.

## 3. Add what they establish

- Additions and corrections only: cited facts, dated history lines, the end of

a thread, contact details exactly as the source writes them.

- What is already right stays as written. Do not reorganise, merge pages or

draw conclusions across pages — that is `rem-muse`. The lead changes only

when a new fact changes what is live; even then keep its quotes.

- Never delete or swap out a quote, an owner instruction or a cited fact

because you found a different one; add yours beside it. Remove a sentence

only when a source contradicts it, and say which.

- Cite with the id on the material's `Source:` line (`gmail:` or `outlook:`

plus 12 hex characters), never a provider's raw message id or thread id.

- The owner is `the user`, never their name, even when mail is signed with it.

- A link that only confirms what the page says is cited once, not quoted.

- A web page about someone with the same name is not this person unless

something ties them (the same address, company, photo caption, link from

their mail).

- A personal contact detail found on the web (a mobile or WhatsApp number, a

home address, a private email) goes on the page only if the subject

published it themselves for people to contact them (their own site or

profile's contact section), and it ends with `[personal]` before its claim

number: `+61 4xx xxx xxx (WhatsApp, on their site for contact) [personal]

[9]`. A number only on a company page is the company's: qualify it so.

A number the subject publishes on their own site or profile is theirs even

when a company page also lists it: write it once, as theirs, with

`[personal]`, and cite both. Anything else found about their private life

stays off the page.

- Anything you add to `Uncertainties` is a question ending in `?`.

Keep the page near the page rules' size (about 15,000 characters). As it

fills, add only what changes Facts, the lead, `Open threads` or `History`;

other findings go in your reply as followed leads, so muse can use them.

Before you finish, compare your page with the page you were given: every

`“…”` quote and every `Sources` id it had must still be on the page, and every

`[n]` in the text must be defined under `Sources`. Restore any you dropped.

+ A dropped quote or source id is refused and sent back for repair.

## Final reply

After the edit, list every lead, one line each:

```

- followed | <lead> | <where it came from [n]> | <what it gave, or "nothing">

- open | <lead> | <where it came from [n]> | <why it matters> | <how to follow it>

- blocked | <lead> | <where it came from [n]> | <what stopped you: login, paywall, gone>

```

Open leads are the next run's work and muse's material; be specific: not

"look into his company", but "read the 2025-11 thread `Re: SOW v2`

referenced in [7]; the page has no reply".