deal-intake
Skill0 starsChanged 59 days agoMITView on GitHub
Record a document that arrived mid-negotiation into the deal file — a landlord's email giving a new owner or notice address, legal's answer to a question the lease left open, finance's rent roll, a consent letter, an assignment nobody had. Use when something arrives after the audit and the deal file has to reflect it, when asked to update a deal file from an email or an attachment, or when a browser form would be the alternative. The agent counterpart to intake.html and review.html; it does not re-audit and it does not negotiate.
In this repository4 other files it gives its agents.
- amendment-draftingSkill
- lease-auditSkill
- negotiation-playbookSkill
- redline-reviewSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 37
If you cannot tell which it is, ask.
- Very specific: line 79
If none of those fire, the chain entry is probably classified wrong.
- Very specific: line 100
If legal's answer contradicts what the audit read, that is a correction and it belongs in audit.corrections[] with legal's answer as the …
- Distinctive: line 119
Only where the ledger proves the audit misread an instrument does a correction follow — and then the reason cites the instrument, not the …
- Distinctive: line 185
Every path you wrote, classified — missing, corrected, or changed.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
Deal intake
Something arrived. Put it in the deal file, correctly, so that everything downstream sees it.
This exists because the toolkit's data entry lives in browser forms —
intake.html, review.html — and an agent cannot click them. That is fine
right up until it becomes the way the discipline gets bypassed: a form that
asks "did this CHANGE or was it WRONG?" and an agent that silently overwrites
the value are not doing the same job, and the agent's version is the one that
loses the record.
So the rule is: whatever the form refuses to do silently, you refuse to do silently. The forms are the specification, not the competition.
The three kinds of edit, which is the whole skill
A deal file changes in three ways and only two of them are errors.
| What happened | Where it goes | |
| MISSING | The audit never had it. There was no previous answer. | Write the value. No record needed — there is nothing to explain. |
| WRONG | The audit misread the documents. | Write the value and an audit.corrections[] entry. |
| CHANGED | The audit read it correctly, and then the world moved. | Write the value and a changes[] entry. |
Getting this wrong is not a filing error. Recording a change as a correction asserts an error the auditor did not make, and it throws away the only date that matters: a correction is dated when somebody noticed; a change is dated when it took effect, and that is the day a cure period, an option deadline or a notice window runs from.
If you cannot tell which it is, ask. "The landlord's address is different from what the audit recorded" is ambiguous — it is a change if they moved and a correction if the audit misread the letterhead, and the two produce different files.
What arrives, and where it goes
A landlord's email: new ownership, new notice address
The commonest one, and the one with the most downstream consequences.
- Is there an instrument? An assignment, a deed, a merger certificate, a
payee direction. If the email refers to one, the instrument is what goes
in
documentChain[]. If the email is the only evidence, it goes in askind: "correspondence"— and say so in the note, because an ownership claim evidenced only by the claimant's own email is worth less than a recorded instrument and a later reader must be able to see that. - Add the chain entry, with
addedAfterAudit: trueandlearnedOntoday.dateis the date on the instrument, not the date of the email. Re-derivechronologicalOrdinalacross the whole chain afterwards: a found instrument rarely belongs last, and the ordinals are what the recitals read. - Update
site.landlord—name,address,id,entityType,capacity— from current title or corporate records, never from the lease. The lease names who it named in 2004. - Record each field in
changes[]witheffectiveset to the transfer date andinstrumentset to the chain id. A party change with no instrument behind it is an assertion; it is not always wrong, but the absence should be visible.
Three things then happen on their own, and you should check that they did:
- The drafter emits a succession recital, and
alCheckSuccessionblocks export if it does not. notices.addressesbecomes a required clause — every cure period runs off an address, and notice to a party that no longer holds the interest is notice to nobody.peReconcilewill check the new name against the document at stage 7, so a draft in circulation naming the old party will be caught before signature.
If none of those fire, the chain entry is probably classified wrong. kind
does that work: assignment triggers them, correspondence does not.
Legal answers something the lease left open
Defined terms, which section governs which subject, negotiated drafting values.
Almost always MISSING rather than changed — nobody had an answer and now
somebody does — so no changes[] entry.
definedTerms.<Name>takesvalue,definedIn(where the lease says it) andredefinedHerewhen the amendment will introduce it. A term withredefinedHere: trueand an emptydefinedInis a term this instrument is creating, which is a decision worth being deliberate about.sectionMap.<subject>is the section of this lease governing that subject. Clauses that replace a section render it directly into the operative text, so a wrong entry produces an amendment that edits the wrong provision.amendmentTerms.*are what the parties agreed, as against what the lease says today. Insurance limits and money take words-then-figures; the co-location share takes the number alone, because the clause supplies the word "percent" itself.
If legal's answer contradicts what the audit read, that is a correction and
it belongs in audit.corrections[] with legal's answer as the reason.
Finance sends the rent roll
Do not overwrite rent.reconciled or rent.components[]. This is the one
that looks like data entry and is not.
The documents say what is owed. The ledger says what was paid. They disagree more often than not, and the disagreement is the finding — a rent roll that differs from the instruments means an underpayment, an unapplied escalation, or a misread document, and which one it is cannot be settled by overwriting either side.
So a rent roll goes into rent.figures[] as {source: "books", amount, basis,
asOf, note}, beside whatever the audit recorded as source: "audited". The
financial model reads both and reports the variance with a cause where one can
be established, and as an open question where it cannot.
Only where the ledger proves the audit misread an instrument does a correction follow — and then the reason cites the instrument, not the ledger.
A consent letter, an estoppel, a side agreement
Straight into documentChain[] with the right kind, plus audit.sources[]
with documentRef pointing at the chain id and an honest extraction and
legibility. That link is what lets the review put findings read from a poor
scan at the top of the queue; without it the field reaches nobody.
A found instrument may also mean the audit is now incomplete rather than merely out of date — a consent can move an equipment right, an estoppel can contradict an option count. Say so, and recommend a re-audit with the current file as the baseline where it matters.
Hard rules
These are not style. Each one exists because breaking it produces a file that looks right.
Never overwrite a value the audit established without recording which kind of
edit it was. site.*, term.*, rent.*, options.*, definedTerms.* where
the audit filled them, and audit.terms.*. If you are writing over something,
say why in the file.
effective is not recordedOn. The date it took effect and the date you
learned of it are different facts and the schema keeps both.
Keep what you replaced. was on every correction and change. What the lease
said before is not superseded by what it says now — it is the record of what
earlier instruments were signed against, and on the day the lease is disputed
that is evidence.
Correcting the same field twice keeps the ORIGINAL was. The second edit
replaces your first answer, not the audit's reading.
Validate before you hand the file back. src/lease-validate.js is the
authority; Lease-Proposal-Comparator.html is the only page that runs it on
load, so a human opening the file there is the cheapest check. An unknown key
is a warning and a malformed enum or date is an error, and an error means the
numbers would change.
Never write a deal file into the repository. They describe real sites and
name real parties. deals/ is gitignored; anywhere outside the repo is fine.
Mark confidential: true.
What you cannot do here
You are not re-auditing. If a found instrument changes the rent history, the
term arithmetic or the options, stop and say so — that is lease-audit
again, with this file as the baseline, and it is a different job with different
care.
You are not negotiating. amendmentTerms records what the parties agreed; it
does not decide it.
You are not deciding whether a party change is genuine. You record what the document says and whose document it is. Whether a claimed assignment is effective is a title question, and the file should make it easy for somebody to ask it rather than answer it silently.
When you are done
Report, in this order:
- What arrived and what kind of document it is.
- Every path you wrote, classified — missing, corrected, or changed.
- Anything now required downstream: a succession recital, the notices clause, a re-audit.
- Anything you could not settle, and what would settle it.
A list of fields you updated is not a report. The reader needs to know what follows from it.
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.

