Reads a user's Link financial data — transactions, balances, and wallet sources — so agents can answer questions about spending and available source capabilities. Use when the user says "check my balance", "how much did I spend", "show my transactions", "what accounts are connected", "summarize my s

Financial insights Use this skill to answer questions about a user’s Link connected financial data, including: Recent transactions Spending patterns Account balances Linked wallet sources Basic summaries derived from the user’s financial data All commands are read only. They do not move money, initiate payments, modify accounts, or expose payment credentials. Safety and privacy Do not retrieve financial data until the user is authenticated with the required source actions. Only retrieve the data needed to answer the user’s request. Do not run every list command by default. Do not expose sensitive identifiers, access tokens, credentials, or payment instrument details. Summarize financial information at the level needed to answer the user’s question. If the user asks for an action that would move money, reference skills/create payment credential/SKILL.md instead. Authentication Before retrieving financial data, check whether the user is authenticated and whether the current session has the required source actions. When present, inspect authorization details in the response for entries with type: "source" and the required actions. The field may be absent when the token endpoint did not return authorization details or when authentication comes from LINK ACCESS TOKEN ; in that case, run only the minimum data command needed and handle a permission error as described below. If the user is not authenticated, start a login that requests only the source actions needed for the requested data. If the user is already authenticated but one or more required source actions are missing, use auth upgrade instead of auth login . auth upgrade preserves the current session while the user approves the additional access and replaces it only after approval succeeds. Use the minimum required source actions: Transactions processed through Link: read link transactions Transactions imported from bank connections: read external transactions Account balances: read balances Data source details and descriptions: read source details If the user asks a question that requires multiple data types, request all relevant actions together. Example for a new login that needs all financial data types: Example for adding balance access to an existing session: Replace <your agent name with a clear name for the agent or application. Present the returned verification url to the user, then follow the response's next instruction or poll with: Do not proceed until authentication or the access upgrade succeeds. If the approval expires, is denied, or times out, report that outcome instead of repeatedly starting new authorization flows. Choosing the right command Use the smallest command set that answers the user’s question. User asks about Command Recent purchases, merchants, spend, transaction history, income, deposits, subscriptions link cli transactions list Current available balance, account balance, cash position link cli balances list Connected accounts, cards, banks, wallet sources, source metadata link cli sources list Examples: “How much did I spend on restaurants last month?” → Use transactions only. “What is my current checking account balance?” → Use balances only. “Which accounts are connected?” → Use sources only. “Summarize my cash position and recent spending.” → Use balances and transactions. Output format Use JSON for agent readable structured output. The default toon format is intended for humans. Prefer format json whenever parsing, filtering, aggregating, or summarizing results. All monetary amounts across all endpoints are integers in the currency's smallest unit (e.g. 152340 = $1,523.40 USD). Format amounts with a currency aware formatter that uses the currency's ISO 4217 minor unit exponent; do not assume every currency has two decimal places or always divide by 100. Keep sign interpretation field specific. Only transactions.amount uses negative for money leaving the account and positive for money entering it. Do not apply transaction sign semantics to balance fields; interpret current , cash.available , and credit.used according to the balance type. Sources (concept) A source is a financial account connected to the user's Link wallet — a bank account, credit card, savings account, etc. Each source has a unique id (e.g. csmrpd abc123 ) that other endpoints may expose as source id : In transactions list , source id indicates which account a transaction belongs to. In balances list , each balance entry includes a source id identifying the account. In sources list , the full source metadata (name, institution, type, status) is returned. Use a source id to correlate data across commands — for example, to find transactions for a specific account or to match a balance to its source type. Do not assign transactions with a null source id to a source by guessing from their description. Transactions Use transactions to answer questions about spending, income, merchants, categories, recurring payments, deposits, or account activity. Common options: Flag Description start date Only transactions on or after this date (YYYY MM DD). end date Only transactions on or before this date (YYYY MM DD). category Filter by category. origin Filter by origin: link or external connection . source Filter by source ID (repeatable). See [Pagination]( pagination) for shared list controls. Response fields Field Note amount Negative = money leaving the account (debit/purchase), positive = money entering (credit/deposit). origin external connection (from linked bank/card) or link (Link native transaction). category May be null if unclassified. status API provided status string. Do not assume a closed set of values; observed values include succeeded . Interpret or filter a status only when its meaning is known. For transaction summaries: Normalize signs consistently before calculating totals. Distinguish debits from credits when possible. Group by merchant, category, account, currency, or time period only when relevant. Mention if the answer is based on a limited retrieved window. Balances Use balances to answer questions about current account balances or available funds. Flag Description source Filter by source ID (repeatable). See [Pagination]( pagination) for shared list controls. Response fields Field Note type cash (bank/savings) or credit (credit card/line of credit). Determines which sub object is present. current Balance before pending transactions. Not the same as available funds. cash.available Object mapping currency codes to available funds (current minus outbound pending plus inbound pending). Only present when type is cash . credit.used Object mapping currency codes to credit used. Only present when type is credit . as of When the balance was last updated — may be stale by hours or days. When summarizing balances: Preserve currencies. Do not add balances across different currencies unless the user explicitly asks and exchange rate data is available. Use the current field as the default definition of a balance, unless the user's question requires considering pending transactions. If multiple sources are returned, summarize by account/source. Sources Use sources to answer questions about connected wallet sources, linked accounts, or available financial data sources. See [Pagination]( pagination) for shared list controls. Response fields Field Description id Unique source identifier (same as source id in other endpoints). name Display name of the source. type Source type (e.g. card , bank account ). capabilities Object indicating what data is available. Each key (e.g. balances , transactions ) maps to an object with a status field (e.g. eligible ). external connection.status Connection status to the external institution. granted actions List of actions the user has granted for this source. When summarizing sources: Include only non sensitive metadata needed for the answer. Avoid exposing full account numbers, credentials, tokens, or payment instrument details. Prefer labels such as institution, account type, source status, and last updated time when available. Pagination All three list commands support the same pagination flags: Flag Description limit Maximum results per page (1 100). Prefer 100 when multiple pages may be needed. starting after Fetch the next page after a cursor value. ending before Fetch the previous page before a cursor value. Use for reverse navigation, not normal forward collection. JSON responses contain a data array and may contain has more . They do not provide a separate next cursor field. When has more is true , derive the next cursor from the final item in data : Command Next cursor transactions list Final transaction's id . balances list Final balance's source id . sources list Final source's id . For example: Keep all filters identical across pages and change only starting after . Stop when has more is false or absent, or when enough data has been retrieved for a non exhaustive lookup. If has more is true but data is empty or the required cursor is null or missing, stop and report that pagination could not continue. Do not exhaustively paginate unless the user’s request requires a complete bounded result, such as a total for a specified time range. Answering user questions When answering: State the direct answer first. Mention the relevant time range and data source. Note any limitations, such as partial pagination, missing categories, pending transactions, or unsupported currencies. Avoid dumping raw records and object IDs unless the user asks for them. Prefer concise summaries, totals, and notable patterns. Example response style: Error handling If authentication fails, ask the user to re authenticate. If a command returns no data, say that no matching Link financial data was available for the requested scope. If the CLI returns an error indicating missing permissions or source actions, request only the specific missing action. Use auth upgrade when a session is already authenticated and auth login when it is not, then wait for approval before retrying the data command once. If data is incomplete or paginated, clearly state that the answer is based on the data retrieved so far. Guardrails Do not: Move money. Initiate payments. Modify financial sources. Retrieve unrelated financial data. Request broader source actions than needed. Expose credentials, tokens, or full payment details. Present uncertain derived insights as definitive. Do: Use read only commands. Authenticate before retrieval. Request the minimum required source actions. Use format json for parsing. Retrieve only the data needed. Summarize clearly and note limitations.

financial-insights