agentleFS
McGuireKevinMcGuireKevin/cell-site-lease-negotiation-toolkit/.claude/skills/amendment-drafting/SKILL.md

amendment-drafting

Skill0 starsChanged 59 days agoMITView on GitHub

Prepare a telecom site lease amendment the way the carrier-side primer requires — assemble and conform the document stack, trace parties and authority, fix the term arithmetic, build the consent list and the issue list, choose the instrument, brief the drafter, review every counterparty draft against the term sheet, and run the pre-execution and post-closing checklists. Use after the playbook and before draft.html, and again on every returned draft. Stage 5.5 through 7 of the lease negotiation pipeline.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    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
  1. Distinctive:

    All three, every time the skill runs before drafting.

    line 20
  2. Distinctive:

    Read the entire stack before writing a word: Every instrument, consent, drawing set, estoppel and side letter — from the DMS and the lease …

    line 38
  3. Distinctive:

    Never redefine a term the lease already defines: Add a distinct new name. "Effective Date" is the usual collision.

    line 49
  4. Very specific:

    Do not mark an item done because the deal file has a value.

    line 85
  5. Distinctive:

    Read definitions, boilerplate, ratification, conflict, integration and exhibits with the same care as rent and term.

    line 176
  6. Very specific:

    Every new request becomes a redline.html disposition.

    line 180
  7. Distinctive:

    Nothing checks that the conformed copy is current: when the next amendment starts.

    line 257

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Amendment drafting

The playbook says what to ask for. The drafter assembles clauses. Neither one reads the whole stack, traces who actually holds the interests, or checks that the instrument being produced is the right instrument. This skill does, following docs/amendment-primer.md, which is the tracked distillation of the carrier-side drafting primer. Read that file first; the numbered references below (1.1, 4.3, 5.4) are to its sections.

OutputWhat it is
amendmentPlan block in <site>.deal.jsonStructured: intake, term summary, consents, issue list, instrument choice, QC, post-closing.
<site>-amendment-plan.mdThe brief a drafter and a reviewer read. Prose with the checklists filled in.
<site>-amendment-rider.mdThe sections of the instrument the drafter cannot emit — written here for pasting into the .docx in Word. Produced whenever the drafting brief calls for one.

All three, every time the skill runs before drafting. On a counterparty draft, the plan and the block are updated; the rider is revised only if the instrument changes.

Before anything else

  • docs/amendment-primer.md — the substance. Appendix B's fourth column says what the toolkit catches and what it does not; the gaps are this skill's job.
  • The deal file. If audit.triage.recommendation is no-action, stop — there is nothing to draft. If there is no playbook block, stop and say the playbook comes first: an amendment implements a deal, it does not negotiate one (5.2).
  • .claude/skills/negotiation-playbook/SKILL.md — for what playbook.points[] and doNotRaise[] mean.
  • clauses/clauses.json and docs/clause-library.md — what the drafter can express.
  • docs/drafting-traps.md — the failure modes already designed out, and the primer pitfalls that are not.

Hard rules

Read the entire stack before writing a word. Every instrument, consent, drawing set, estoppel and side letter — from the DMS and the lease file, not only from the lease-administration system, which holds incomplete sets (Part 2). audit.sources[] with extraction: unavailable is the list of what is still missing, and the plan says whether drafting can proceed without it.

The conformed working copy is internal. Mark it NON-BINDING, never attach it, never reference it in an instrument as controlling (1.12).

Never renumber an existing agreement by amendment. New material goes in as new lettered subsections or lettered sections (1.1).

Never redefine a term the lease already defines. Add a distinct new name. "Effective Date" is the usual collision; the amendment's own date is the Amendment Effective Date (1.7).

Three dates, defined separately. Execution, Amendment Effective Date, and the commencement date of each changed obligation. No "effective as of the date first written above" (1.8).

Ratification is mutual and knowledge-qualified, with the landlord confirming rent current, no default, equipment permitted and no back rent claimed. Exhibits are depictions, never caps (1.9).

Conflict and integration clauses are scoped to the amendment's subject matter. "Lease" is defined by a schedule listing every instrument and every consent to be preserved (1.2, 1.3).

State the term arithmetic in the instrument. Where the term stands, which options remain and are unaffected, what is added, how the added periods work, and that every carrier termination right applies throughout (1.11, 4.3).

Never invoke Section 6409(a) as a reason the landlord must consent. It governs zoning review, not property rights, and using it that way costs credibility (4.9).

Record a memorandum of amendment whenever the term or the premises change or the original was recorded (1.10).

Never write client content into the repository. The plan, the rider and the deal file go to deals/ or outside the repo, with confidential: true.

Procedure

1. Intake (5.1)

Fill amendmentPlan.intake[] — one entry per checklist item with status: done | open | n/a, source (DMS path, lease-admin export, title report, corporate records) and a note. Where the corporate environment supplies it, say which system: the lease-administration export for term status and payment history; the document management system for the stack and prior drafts by round; title or public-records access for vesting, memoranda and later mortgages; Word Compare for the conformed copy.

Do not mark an item done because the deal file has a value. The audit read documents; intake confirms them against the systems of record.

2. Conformed working copy (1.12, 5.3)

Apply every amendment to the base lease in date order and write the result as <site>-conformed-NONBINDING.md (or a .docx produced in Word from the DMS versions with Compare). Every paragraph carries the instrument and section that last touched it. This is the first of the two baselines every counterparty draft is compared against; redline.html supplies the second (the last draft we circulated). Record the file name in amendmentPlan.conformedCopy and mark it non-binding in the document itself.

3. Term summary (1.11)

amendmentPlan.termSummary: current period and its end, options exercised and remaining, whether renewal is automatic, the NNR period and next deadline (or unknown with the reason), any landlord termination or redevelopment right and the periods it reaches, the landlord's own estate where it is a ground lessee or life tenant. Every figure cites term.*, options[] or an audit.terms entry. An unknown stays unknown.

4. Parties and consents (1.6, 1.13)

amendmentPlan.parties: the carrier's current contracting entity and its succession chain; the landlord of record, the fee owner if different, any rent-stream acquirer or easement holder, and the payee — each from title or corporate records, not from the payee in the administration system. Whether a portfolio transaction governs who may sign.

amendmentPlan.consents[]: lender (and whether the SNDA excludes unconsented amendments or prepaid rent), owners' association, ground lessor, co-owners, spousal joinder, tower company, public body, internal delegation of authority. Each with required: yes | no | unknown, basis, and status. The amendment's effectiveness is conditioned on the ones marked yes.

5. Issue list (Part 2, 4.8)

amendmentPlan.issues[] in two groups:

  • Carrier asks — one per playbook.points[] entry that is open or won, carrying the point's rank and the clause that gives effect to it.
  • Expected counterparty asks — from Part 4 by amendment type (equipment 4.1, rent 4.2, extension 4.3, relocation 4.4, assignment 4.5, operational 4.6, site type 4.7) and from the counterparty type (direct landlord, rent-stream acquirer, tower company). Each with the primer trap it carries and the position from playbook.points[].anticipated[] or negotiation/positions.json where one exists.

Say which party needs the amendment and what leverage that creates (Part 1).

6. Instrument (1.12, 4.5)

amendmentPlan.instrument: targeted-amendment | restatement | supplement-amendment | notice-or-acknowledgment, with the reason. A name change on a lease that already permits affiliate and successor assignment is a notice, not an amendment. A restatement is chosen only when the stack is genuinely incoherent, and then the plan lists every provision of the conformed copy that the restatement must carry forward.

7. Drafting brief

amendmentPlan.brief: the clauses to select in draft.html (by id, in the order the instrument should read), the sectionMap entries to confirm against the conformed copy, the definedTerms the amendment must define because the lease never did (with the distinct name chosen), and the amendmentTerms values.

Then the rider — <site>-amendment-rider.md — carrying the Appendix A sections the drafter cannot produce from the library, each written for this instrument:

  1. Caption with both parties' site identifiers.
  2. Preamble with current legal names, entity types, jurisdictions and capacities.
  3. Recitals: every prior instrument by title, date, parties and recording information; every assignment and succession; a neutral purpose.
  4. Definitions and incorporation: "Lease" by Schedule 1; the Amendment Effective Date; every new term.
  5. Term arithmetic (where the drafter's term clause hardcodes a different count or notice period, the rider states the correct figures and the clause is edited in Word).
  6. Ratification, confirmations and reservation of rights — mutual, knowledge-qualified.
  7. Scoped conflict clause; no-implication clause naming the preserved rights; non-novation.
  8. Consents and conditions to effectiveness.
  9. Authority representations.
  10. Recording — memorandum of amendment, who executes, records and pays.
  11. Dated execution lines and capacity in each signature block.
  12. Schedule 1 and exhibit legends ("depicts current installation and does not limit …").

Mark each rider section with the drafter section it follows or replaces, so the paste is mechanical.

8. Counterparty draft review (4.8, 5.3)

On every returned draft, before or alongside redline.html:

  • Compare against both baselines: the conformed copy and the last draft we sent.
  • Map every provision of the returned draft to an item on the issue list. Anything that maps to nothing is a new request and goes into amendmentPlan.issues[] as origin: counterparty, with the Appendix B pitfall number it matches.
  • Read definitions, boilerplate, ratification, conflict, integration and exhibits with the same care as rent and term; this is where unrequested changes are placed.
  • Where the form is unworkable, say so and substitute ours.

Every new request becomes a redline.html disposition; the plan records which pitfall each one was.

9. Pre-execution QC (5.4)

amendmentPlan.qc[] — the seventeen items, each {item, status, evidence, note} where status is pass | fail | n/a | open.

Use the item ids from src/amendment-plan.js, not your own wording. The checklist text lives there and comes from the primer; the deal file records answers against ids. An id that file does not carry is reported at stage 8 as an unrecognised check rather than counted, because the usual cause is a typo and a typo is a check that silently never happened.

report.html now reads this. A fail blocks signature there as well as here, and the package reports how many items are still open — an untouched checklist is not a clean one. Stage 8 also cross-examines: where an item is marked pass and the drafting review is still reporting the thing it asserts, the package says so. Mark pass only against evidence; a pass with an empty evidence raises a validation warning.

10. Execution and post-closing (5.5)

amendmentPlan.postClosing[]: memorandum recorded (number); payment direction and tax documentation delivered; lease-administration entries made (new rent components, commencement dates, escalation dates, notice windows, removal obligations, consent conditions); notice addresses updated in the system; copies delivered to every consenting party; SNDA or title endorsement obtained; conformed copy updated; executed amendment filed with the stack.

Emit

The plan follows the order a reviewer needs:

  1. What is being amended and why, who needs it, and the instrument chosen
  2. The stack — what was found, what is missing, and whether drafting can proceed
  3. Parties, authority and consents
  4. Term arithmetic
  5. Issue list — our asks, their expected asks
  6. Drafting brief and the rider
  7. On a revision: what the counterparty changed, mapped to the issue list, with pitfall numbers
  8. QC and post-closing status

Before you finish

  • Every intake item has a status and a source system
  • The conformed copy exists, is named, and is marked non-binding
  • Term arithmetic cites the deal file; unknown stays unknown
  • Parties traced from title and corporate records, not from the payee
  • Consent list complete; effectiveness conditioned on the ones required
  • Every playbook point is an issue; every expected counterparty ask carries its pitfall number
  • The rider covers every Appendix A section the drafter cannot produce
  • No "Effective Date" redefinition; three dates defined separately
  • Conflict and integration scoped; ratification mutual
  • confidential: true; nothing written into the repository

Not built yet

Three things listed here have since been built, and the entries are kept as a record of what each one replaced.

A conformed-copy builder — conformed.html and src/conform-build.js. It applies each amendment in date order from documentChain and marks the result non-binding. It refuses narrative instructions and exhibit replacements rather than guessing, reproducing them verbatim as unresolved, so the copy is honest about what a person still has to do. amendmentPlan.conformedCopy carries the file name.

The rider — src/amendment-skeleton.js. The drafter now emits the caption, the preamble with entity types and capacities, the chain recital from documentChain, the succession recital, Schedule 1, mutual knowledge-qualified ratification, a scoped conflict clause, no-implication, non-novation, authority representations, recording and dated execution lines. The rider is now the site-specific residue it was always meant to be.

Term-sheet mapping — src/term-sheet-map.js, in redline.html. It runs in both directions: a paragraph they added that answers to nothing agreed, and a paragraph of ours that did not come back. It reads the aligner's own classification rather than scoring the text, so it agrees with the rest of the stage about what counts as the same provision.

Still outstanding:

Nothing checks that the conformed copy is current when the next amendment starts. The plan records the file name; it cannot tell whether it predates the last instrument in the chain.

amendmentPlan.issues[] and deviations[] are not keyed to each other. The term-sheet map works from playbook.points[] and the clause selection. An unmapped provision is a mechanical finding, but tying it back to the issue list — so a conceded issue shows as conceded in both places — is still done by reading.

What changes after the plan is written

The plan is built once and the negotiation runs for months. Three blocks in the deal file move underneath it, and each one has a drafting consequence rather than merely a record-keeping one.

changes[] — the counterparty stopped being the same counterparty. §4 reads title and corporate records at intake. Where the landlord is assigned, merges, renames or moves its notice address after that, the fact lands in changes[] (via deal-intake), not in a re-run of §4 — and it is the one kind of update that rewrites the instrument rather than the plan:

  • a succession recital becomes required, and alCheckSuccession blocks export without it;
  • notices.addresses becomes a required clause, because every cure period runs off an address and notice to a party that no longer holds the interest is notice to nobody;
  • the name appears in the preamble, the recitals, the notices clause, the signature block and the running head, and correcting some of those and not the rest is the common failure — which is what peReconcile catches at stage 7, after the draft is already in circulation.

So re-read changes[] at every turn, not only at intake. A party change dated mid-negotiation also means the consents[] list was built against the wrong party: a lender consent obtained from the assignor's lender is not the consent the amendment is conditioned on.

redline.openStubs[] — the blanks they left. On counterparty paper a draft arrives with ruled lines, bracketed prompts and unfilled merge fields where the negotiation has not landed. Every signal in the redline reads what the text says, so a blank is invisible to all of them. Treat an open stub as a term nobody has agreed, not a formatting defect: it is either something the other side completes after signature or a gap a court construes. It belongs in the QC checklist as an open item, and report.html will not produce a package while any remain.

redline.reconciliation — the document and the file disagree. Parties, notice address, site id, agreed rent, term end, effective date. These are not concessions to weigh and not issues for the issue list. The fix is upstream: deal-intake records the change and the drafter reissues.

audit.corrections[] — the reading you built on may have moved. The term summary in §3 is arithmetic over what the audit read. Where stage 1.5 corrected a date, a rent figure or an option count, the summary inherits the error until it is recomputed. Check audit.reviewedOn against the date on your plan before trusting §3.

Data entry is not your job. Where any of this has to be written into the deal file — a landlord's email, legal's answer to an open question, finance's rent roll — that is deal-intake, which classifies the edit as missing, wrong or changed and keeps what it replaced. Recording a change as a correction asserts an error the auditor did not make and throws away the effective date, which is the date the cure period and the notice window run from.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.