--- oa_review_kit: v1 guide_slug: cross-border-tax-workflow-base guide_version: cross-border-tax-workflow-base@2026-05-31T18:20:05.606Z archetype: other --- # Review kit: Cross Border Tax Workflow Base Thank you for reviewing this Guide. This kit is one file with three parts: how to use it, an interview prompt for your AI, and the Guide itself. ## How to use this kit (3 steps, about 15 minutes) 1. Open the AI you already use (ChatGPT, Claude, Gemini, anything that reads markdown) and paste in everything from "INTERVIEW PROMPT" below, including the Guide at the end. 2. Your AI interviews you like a colleague, one question at a time. Just talk: war stories, walk-throughs, the mistakes you catch. No writing required. 3. Your AI writes your answers up as a single markdown file. Hand it back at openaccountants.com/skills/cross-border-tax-workflow-base/handback (also linked from the Guide page: "Hand back your file"). What you added is published under your name and credential. If your AI cannot produce the exact output format, hand back whatever you have: a revised Guide file, a worksheet, or plain notes. We take those too, and a person reviews them by hand. The format below is the one we can apply straight away. --- # INTERVIEW PROMPT (paste from here down into your AI) You are interviewing a practising accountant about how they actually do the work covered by the attached Guide ("Cross Border Tax Workflow Base", slug `cross-border-tax-workflow-base`). Interview them like a colleague doing a handover. Do not lecture. Ask ONE question at a time and wait for the answer. Chase war stories and specifics: what kind of client, which portal step, how big the penalty was. The rates, thresholds, and citations are our job; we refresh those from primary sources. Capture ONLY what is NOT derivable from law: - order of operations, and what a wrong order corrupts - what to ask a client before computing anything - what to assume when a fact is unknown, and how it gets flagged - the most-missed traps, with penalty size and who falls in - how the portal or filing channel actually behaves - what has to reconcile before anyone signs - when to refuse the work and hand it to a human specialist If the accountant corrects a rate, threshold, or deadline in the Guide along the way, record it in the FACT CORRECTIONS table, but do not steer the interview toward numbers. ## Questions to work through Ask these in order, one at a time. Skip any the accountant has already covered; follow up where a story has specifics worth pinning down. Each question is tagged with the method slot(s) it feeds. 1. [sequence] Walk me through the last one of these you did for a real client, start to finish. What did you open first, and why that order? 2. [intake_questions] A new client sits down for this work. What are your first five questions before you touch a number? 3. [evidence] Which documents do you insist on seeing, and which do you take the client's word for? 4. [trap] When you review this work drafted by someone else, what mistake do you catch most often? 5. [conservative_default] When a key fact is unknowable at draft time, what do you assume, and how do you flag it? 6. [judgment_rule] When the law allows two routes, how do you actually pick, and what do you write down about the choice? 7. [cross_check] Before you sign, what has to reconcile with what, and how close is close enough? 8. [filing_mechanics] Walk me through the actual submission: the portal steps, the order things must happen in, what locks, what you can't undo. 9. [scope_gate] Which clients do you refuse or refer to a specialist for this work? What makes you stop? 10. [unsettled_law] Anything here you deliberately won't finalise right now because the rules are moving? 11. [handback_protocol] What exactly do you hand over at the end? What's in your working paper? ## Method slots (for tagging the write-up) - `scope_gate` (Scope gate and refusals): when to stop and send the client to a human - `sequence` (Order of operations): what order to do things in, and what a wrong order corrupts - `intake_questions` (Client intake questions): what to ask a client before computing - `evidence` (Documents and evidence): which documents to insist on, and what is draft-grade vs file-grade - `judgment_rule` (Judgment rules): how a practitioner actually picks when the law allows two routes - `conservative_default` (Conservative defaults): what to assume when a fact is unknowable at draft time - `trap` (Traps and most-missed items): the mistakes everyone makes, what they cost, and who falls in - `filing_mechanics` (Portal and filing mechanics): how submission actually works: channel, order, what locks - `cross_check` (Cross-checks before signing): what has to reconcile with what before delivery, and how close is close enough - `pattern_library` (Pattern library): how messy real-world data (bank lines, payout platforms) maps to tax categories - `edge_case` (Edge-case playbook): the client situations that change the method, not just the numbers - `unsettled_law` (Unsettled-law flags): what not to finalise right now, and why - `handback_protocol` (Hand-back protocol): what the finished working paper contains and who reviews it ## Output format: oa-handback v1 When the interview is done, write the answers up as ONE markdown file in exactly this shape. Fill in the reviewer's real name, credential, and email (ask for them at the end if they have not come up). Every method block gets a `### [method:]` heading where `` is one of the 13 slot ids above. Keep `guide_slug` and `guide_version` exactly as given. Omit any section the interview produced nothing for, but keep the headings that remain exactly as shown. The `fact_key` column may be left blank when unknown. ```markdown --- oa_handback: v1 guide_slug: cross-border-tax-workflow-base guide_version: cross-border-tax-workflow-base@2026-05-31T18:20:05.606Z reviewer_name: reviewer_credential: # free text: CPA, EA, ACCA, Steuerberater... reviewer_email: verdict: --- ## METHOD ### [method:filing_mechanics] ### [method:intake_questions] - - ... ## FACT CORRECTIONS | fact_key | current | correct | source | |---|---|---|---| | | | | | ## FLAGS - [unsettled] - [refer] ## NOTES ``` If for any reason you cannot produce this exact format, output the accountant's corrections and methods as clear plain notes instead. The hand-back page accepts plain notes and revised Guide files too; this format is an optimization, never a gate. --- # THE GUIDE UNDER REVIEW --- name: cross-border-tax-workflow-base description: Foundation workflow base for cross-border / international personal-tax content skills. Contains the residency-map intake, the sequenced-plan output contract (a cross-border answer is an ORDERED set of steps, not N separate answers), the cross-border conservative-default principle, the AUDIT FLASH POINT marker convention, the double-tax-relief / treaty-bridge convention, and the mandatory human hand-off. This skill provides workflow architecture only — it contains no country-specific or topic-specific rules. It MUST be loaded alongside a topic content skill (e.g. us-feie-ftc, us-foreign-trust-reporting) and the relevant country skills. This base is the foundation every international content skill loads on top of. jurisdiction: GLOBAL domain: cross-border tax_year: 2025 --- # cross-border-tax-workflow-base ## Cross-Border Tax Workflow Base v0.1 ## What this file is **This file is the foundation workflow base that every cross-border / international content skill loads on top of.** It is the shared contract. It carries no country rules and no topic rules of its own — those live in the topic content skills (FEIE/FTC, FBAR/FATCA, CFC/GILTI, foreign trusts, exit tax) and in each country's own skills. A content skill is useless without this base, and this base computes nothing without content skills. Load both, plus the country skills the router identified. **Currency.** Content skills state their own tax-year currency. This base is year-agnostic and does not expire. **This output is a working paper handed to a licensed accountant — never a filed return.** Cross-border situations carry the highest penalty exposure in tax (FBAR, 3520, 8938, exit tax), and the order of events is often irreversible once executed. The skill produces a sequenced plan, a working paper, and a reviewer brief — and it ends by handing that working paper to the named accountant for the lead country **before** the person acts. ## Section 1 — The sequenced-plan output contract A cross-border answer is **not** a stack of single-country answers. It is an **ordered plan**. Every cross-border content skill produces both layers below. ### Layer A — Reference layer (the rules) - **Layer A — Reference layer (the rules)** — For each country and topic engaged, a faithful, citation-anchored statement of the rule (statute / form / treaty article), expressed as decision trees an agent can reason over. Answers: "What does each system require, and where is it written?" Every rule cites its source (e.g. IRC §911, §877A, Form 3520, the relevant treaty article). _(IRC §911; §877A; Form 3520)_ ### Layer B — Executable layer (the sequenced plan) - **Layer B — Executable layer (the sequenced plan)** — A step-by-step procedure that takes the person's facts and the order of events fixed by the router, and produces an ordered list of steps. Each step states: 1. What happens (the event, and when). 2. Which country taxes it, on what basis (citizenship / residence / domicile / source), citing the Layer A rule. 3. How double tax is relieved at that step — the treaty bridge (FTC, exemption, re-sourcing, tie-breaker), citing the article/section. 4. Which forms / filings the step triggers in each country (with deadlines). 5. Why this step is ordered where it is — what breaks if it moves. Layer B answers: "Given these facts and this order, who taxes what, in what sequence, and what gets filed?" **Rule:** every Layer B step must reference the Layer A rule it executes. No position without a cited source — and no treaty benefit asserted without the article and the filing that claims it. _(Layer B rule statement)_ ## Section 2 — The double-tax-relief / treaty-bridge convention The single most valuable part of a cross-border answer is the bridge: how the two (or more) systems' taxing rights are reconciled so the same income is not taxed twice without relief. For every item that more than one country taxes, the skill MUST state, explicitly: - Which country has the primary taxing right and which gives relief. - The mechanism — foreign tax credit (§901/§904), exemption/exclusion (§911), treaty re-sourcing, residence tie-breaker, or a reduced treaty withholding rate. - The filing that actually claims it (e.g. Form 1116, Form 2555, Form 8833) — a relief that is not claimed on the right form is not relief. - The residual double tax, if any, that the mechanism does not eliminate (e.g. rate differentials, the US saving clause, state tax that ignores the treaty). - **The saving clause caveat** — Most US treaties contain a saving clause that lets the US tax its citizens/residents as if the treaty did not exist, subject to listed exceptions. Never assert a treaty benefit for a US person without checking the saving clause and its exceptions first. _(US treaty saving clause)_ ## Section 3 — AUDIT FLASH POINT convention Mark every position that tax authorities (IRS, HMRC, ATO, etc.) actively challenge with a bold marker: > **⚑ AUDIT FLASH POINT —** [the judgement at issue, why it is contested, what evidence supports the position taken, and the filing/disclosure that defuses it.] These are the positions a reviewer must personally own. Surface every flash point a situation triggers in the reviewer brief — never bury them. Typical cross-border flash-point families: residency-cessation date and day-count evidence; treaty-residence tie-breakers; grantor-vs-non-grantor trust characterization; FEIE-vs-FTC elections and the §911(d)(1) tax-home/bona-fide test; covered-expatriate status and exit-tax valuations; PFIC exposure inside foreign funds; reasonable-cause positions for late FBAR/3520/8938; and any position that depends on facts in another country the preparer cannot verify. ## Section 4 — Conservative-default principle (cross-border edition) When facts are incomplete and a position could go either way across a border, the skill takes the position that is harder to challenge and creates the reporting trail, states it has done so, and lists the specific fact that would change the answer. Concretely, when unsure: - Assume the broader taxing right applies (e.g. assume US residence/citizenship taxation reaches the item) rather than the narrower one. - Assume the reporting obligation is triggered (file the FBAR / 8938 / 3520) rather than assuming an exemption — the cost of an unnecessary information return is near zero; the cost of a missed one is penalties measured in tens of thousands. - Assume the higher-tax characterization of an entity or instrument until a fact establishes otherwise. It does not silently pick the favourable treatment. Aggressive positions (e.g. a treaty position that overrides the saving clause, a non-filing position) require an affirmative, documented instruction and a named accountant's sign-off — they are never the default. ## Section 5 — Structured intake (the residency map) If facts needed to run Layer B are missing, ask for them in a single structured block before computing — never piecemeal, never guessing. The router's Step 0 map is the intake; restate any gaps here grouped by the decision they unlock: ``` To sequence this I still need: 1. [Fact] — used in [step / rule ref] 2. [Fact] — used in [step / rule ref] ... If you don't have [fact], I will assume [the broader taxing right / the reporting obligation applies] (§4) and flag it for your accountant. ``` ## Section 6 — Universal self-checks Before delivering output, the agent verifies (each content skill adds topic-specific checks on top): - [ ] Both layers produced (a cited rule for every step) - [ ] The answer is an ordered plan, not N independent country answers - [ ] Order of events stated and justified (what breaks if a step moves) - [ ] For every item taxed by >1 country, the treaty bridge is stated with the relieving mechanism and the form that claims it - [ ] Saving clause checked before asserting any treaty benefit for a US person - [ ] Every US information-reporting form the facts trigger is listed (FBAR, 8938, 3520/3520-A, 5471/8865, 8621) — not just the income forms - [ ] Every figure traces to a source input or a cited rule — no unsourced numbers - [ ] Every triggered AUDIT FLASH POINT surfaced in the reviewer brief - [ ] Conservative default applied and flagged wherever facts were incomplete - [ ] Any uncovered country rule / named-treaty read / PFIC / estate item flagged as an open item, not improvised - [ ] Output marked tier 2 (research-verified); the human hand-off (§8) is executed ## Section 7 — Output specification Deliver, in this order: 1. Situation map — citizenship / residence(s) / domicile and the event, one block. 2. The sequenced plan — the ordered Layer B steps (event → who taxes → relief → filings → why ordered here). 3. Reference-layer trace — the rule / form / treaty article behind each step. 4. Double-tax bridge summary — per item: primary right, relief mechanism, claiming form, residual. 5. US reporting-forms checklist — every information return triggered, with deadlines and penalties. 6. Reviewer brief — the judgements taken, every AUDIT FLASH POINT, conservative defaults applied, and the open items the licensed accountant must personally clear. 7. Hand-off — per Section 8. ## Section 8 — Verification tier and the mandatory hand-off Every cross-border output carries the tier 2 (research-verified) mark: drafted from authoritative statutes, forms, and treaties, and awaiting a named, licensed accountant's sign-off. No cross-border skill output is ever presented as accountant-verified on its own, because no human has yet signed the specific plan. **The workflow MUST end by handing the working paper to a human.** Before the person executes any irreversible step (selling, distributing, ceasing residence, expatriating, filing), call `request_accountant_review` with: - `jurisdiction` = the lead country (the one with the primary taxing right or the largest exposure — usually where the person is, or will be, resident, or the US where citizenship taxation dominates), - `scenario` = the situation map, and - `working_paper` = the full sequenced plan above. The tool routes the plan to the lead verifier for that country so the accountant sees the numbers before the call. Tell the user, plainly: this is a working paper, not advice you can act on yet; a licensed accountant in [lead country] will review it and is the one who signs off and (if engaged) files. 0. **Mandatory hand-off to accountant review** — Call request_accountant_review with jurisdiction, scenario, and working_paper before the person executes any irreversible step. ## Section 9 — Disclaimer These skills provide computational and interpretive guidance on cross-border personal taxation. They are not tax or legal advice, not an engagement, and not a filed return. Cross-border outcomes turn on entity-, residence-, and treaty-specific facts and significant judgement, and the order of events frequently changes the result. Always have outputs reviewed and signed by a qualified, licensed accountant in each relevant country before acting on them. Penalty exposure on cross-border information returns is severe; when in doubt, file and ask.