@zktx.io/sui-mcp 0.6.2 → 0.8.0

package/docs/MCP_TOOLS.md

@@ -6,7 +6,7 @@ This document is the MCP API reference. It owns tool contracts, response fields,
 
 It does not own installation steps or user-question playbooks. Installation and client setup live in `docs/MCP_SETUP.md`. User-question flows and response wording live in `docs/AGENT_BEHAVIOR.md`.
 
-Tool names use dot prefixes and avoid arbitrary shell, arbitrary Move calls, and private-key handling.
+Tool names use `domain_verb_object`. A name does not grant authority; input schemas, card visibility and scoped permission checks define the allowed effects. Tools do not expose arbitrary shell commands, arbitrary Move calls or private-key handling.
 
 The local stdio gateway forwards each request through authenticated internal
 HTTP. A request body must fit within 4 MiB, including the entire MCP JSON-RPC
@@ -37,38 +37,38 @@ not grant input permission; a saved card is read through its bound resource.
 
 | Tool | Status | Purpose |
 | --- | --- | --- |
-| `read.get_server_status` | Implemented | Returns package, version, evidence policy, network, runtime, `implementedToolsCount`, resources, prompts, and tool implementation status. |
-| `read.list_supported_protocols` | Implemented | Lists current mainnet protocol surfaces and support levels. |
-| `read.list_deepbook_pools` | Implemented | Lists DeepBook mainnet pools from pinned `@mysten/deepbook-v3` constants. |
-| `read.list_deepbook_tokens` | Implemented | Lists DeepBook mainnet tokens from pinned `@mysten/deepbook-v3` constants. |
-| `read.inspect_deepbook_orderbook` | Implemented | Uses pinned DeepBook SDK read methods over Sui gRPC simulation reads with an internal sender placeholder; `ticks` is capped at 50. |
-| `read.get_deepbook_mid_price` | Implemented | Returns a DeepBook pool mid price snapshot from pinned SDK simulation reads. |
-| `read.get_deepbook_usdc_price_history` | Implemented | Returns DeepBookV3 official Indexer USDC candle evidence for a requested official interval. |
-| `read.get_deepbook_usdc_price_at_time` | Implemented | Returns the DeepBookV3 official Indexer USDC candle for or nearest to one target UTC time. |
-| `read.quote_deepbook_action` | Implemented | Quotes raw integer DeepBook quantities through pinned SDK transaction builders and raw `u64` simulation return values with an internal sender placeholder. |
-| `read.quote_deepbook_display_amount` | Implemented | Converts an explicit display source amount through pinned DeepBook token units, then returns scoped display quote facts plus raw quote evidence. |
-| `read.summarize_deepbook_account_inventory` | Implemented | Summarizes active-account DeepBook BalanceManager inventory through pinned SDK simulation reads. |
-| `read.summarize_wallet_assets` | Implemented | Reads coin balances for an explicit address or the active account through Sui gRPC `client.core.listBalances`; accepts `cursor` for pagination. |
-| `read.classify_wallet_assets` | Implemented | Classifies coin balances for an explicit address or the active account by balanceStatus and coin-balance roles; accepts `cursor` for pagination. |
-| `read.list_settlement_asset_groups` | Implemented | Lists supported settlement asset groups derived from pinned mainnet SDK registries. |
-| `read.summarize_settlement_asset_group_parity` | Implemented | Summarizes direct DeepBook mid-price parity across a supported settlement asset group against a declared measurement reference. |
-| `read.preview_intent_evidence` | Implemented | Builds current wallet and DeepBook evidence for a natural-language settlement intent; not transaction building or signing. |
-| `read.list_review_activity` | Implemented | Lists local Sui MCP review evidence for one account. |
-| `read.summarize_review_funnel` | Implemented | Summarizes local review lifecycle counts, status distribution, and review timing. |
-| `read.get_review_session_detail` | Implemented | Returns stored local evidence for one Sui MCP review session. |
-| `read.inspect_sui_transaction` | Implemented | Looks up one Sui transaction digest and stores normalized facts only when the transaction sender or a returned balance-change owner matches a known local wallet. |
-| `read.scan_sui_account_activity` | Implemented | Runs a bounded GraphQL activity scan for a known or explicit Sui account. |
-| `read.summarize_sui_activity_scan` | Implemented | Runs a bounded Sui activity scan and returns requested-account facts plus deterministic normalized-fact analysis without full details. |
-| `read.scan_sui_function_activity` | Implemented | Runs a bounded GraphQL scan for transactions the account sent that called one full `package::module::function`. |
-| `read.summarize_sui_function_activity_scan` | Implemented | Runs the sent-function activity scan and returns requested-account facts plus deterministic normalized-fact analysis without full details. |
-| `read.summarize_sui_account_activity` | Implemented | Summarizes stored normalized Sui activity facts from local SQLite. |
-| `read.get_account_asset_timeline` | Implemented | Builds stored local account asset net-flow bars from normalized activity facts and optional DeepBook USDC token-denominated candle references. Not held balances or complete wallet history. |
+| `read_get_server_status` | Implemented | Returns package, version, evidence policy, network, runtime, `implementedToolsCount`, resources, prompts, and tool implementation status. |
+| `read_list_supported_protocols` | Implemented | Lists current mainnet protocol surfaces and support levels. |
+| `read_list_deepbook_pools` | Implemented | Lists DeepBook mainnet pools from pinned `@mysten/deepbook-v3` constants. |
+| `read_list_deepbook_tokens` | Implemented | Lists DeepBook mainnet tokens from pinned `@mysten/deepbook-v3` constants. |
+| `read_inspect_deepbook_orderbook` | Implemented | Uses pinned DeepBook SDK read methods over Sui gRPC simulation reads with an internal sender placeholder; `ticks` is capped at 50. |
+| `read_get_deepbook_mid_price` | Implemented | Returns a DeepBook pool mid price snapshot from pinned SDK simulation reads. |
+| `read_get_deepbook_usdc_price_history` | Implemented | Returns DeepBookV3 official Indexer USDC candle evidence for a requested official interval. |
+| `read_get_deepbook_usdc_price_at_time` | Implemented | Returns the DeepBookV3 official Indexer USDC candle for or nearest to one target UTC time. |
+| `read_quote_deepbook_action` | Implemented | Quotes raw integer DeepBook quantities through pinned SDK transaction builders and raw `u64` simulation return values with an internal sender placeholder. |
+| `read_quote_deepbook_display_amount` | Implemented | Converts an explicit display source amount through pinned DeepBook token units, then returns scoped display quote facts plus raw quote evidence. |
+| `read_summarize_deepbook_account_inventory` | Implemented | Summarizes active-account DeepBook BalanceManager inventory through pinned SDK simulation reads. |
+| `read_summarize_wallet_assets` | Implemented | Reads coin balances for an explicit address or the active account through Sui gRPC `client.core.listBalances`; accepts `cursor` for pagination. |
+| `read_classify_wallet_assets` | Implemented | Classifies coin balances for an explicit address or the active account by balanceStatus and coin-balance roles; accepts `cursor` for pagination. |
+| `read_list_settlement_asset_groups` | Implemented | Lists supported settlement asset groups derived from pinned mainnet SDK registries. |
+| `read_summarize_settlement_asset_group_parity` | Implemented | Summarizes direct DeepBook mid-price parity across a supported settlement asset group against a declared measurement reference. |
+| `read_preview_intent_evidence` | Implemented | Builds current wallet and DeepBook evidence for a natural-language settlement intent; not transaction building or signing. |
+| `read_list_review_activity` | Implemented | Lists local Sui MCP review evidence for one account. |
+| `read_summarize_review_funnel` | Implemented | Summarizes local review lifecycle counts, status distribution, and review timing. |
+| `read_get_review_session_detail` | Implemented | Returns stored local evidence for one Sui MCP review session. |
+| `read_inspect_sui_transaction` | Implemented | Looks up one Sui transaction digest and stores normalized facts only when the transaction sender or a returned balance-change owner matches a known local wallet. |
+| `read_scan_sui_account_activity` | Implemented | Runs a bounded GraphQL activity scan for a known or explicit Sui account. |
+| `read_summarize_sui_activity_scan` | Implemented | Runs a bounded Sui activity scan and returns requested-account facts plus deterministic normalized-fact analysis without full details. |
+| `read_scan_sui_function_activity` | Implemented | Runs a bounded GraphQL scan for transactions the account sent that called one full `package::module::function`. |
+| `read_summarize_sui_function_activity_scan` | Implemented | Runs the sent-function activity scan and returns requested-account facts plus deterministic normalized-fact analysis without full details. |
+| `read_summarize_sui_account_activity` | Implemented | Summarizes stored normalized Sui activity facts from local SQLite. |
+| `read_get_account_asset_timeline` | Implemented | Builds stored local account asset net-flow bars from normalized activity facts and optional DeepBook USDC token-denominated candle references. Not held balances or complete wallet history. |
 
 Read-only tools split by address requirement.
 
 Address-free reads do not need wallet connection. Examples include DeepBook pool lists, orderbook snapshots, mid price, and quote facts.
 
-Current asset reads need either an explicit public Sui address supplied by the user or a selected account with a usable current wallet connection. Stored activity and transaction-result reads retain their own read-context rules.
+Current asset reads need either an explicit public Sui address supplied by the user or the saved Sui read address with a usable current wallet connection. Stored activity and transaction-result reads retain their own read-context rules.
 
 An explicit public-address read does not prove ownership, create active account context, or authorize active-account-only tools.
 
@@ -112,15 +112,15 @@ Token list entries are keyed by SDK symbol and include the pool keys that refere
 
 Token entries include `decimals` derived only from pinned `mainnetCoins.scalar` when every scalar is a power of ten. If that invariant fails, the token registry tool fails closed instead of guessing.
 
-`read.get_deepbook_mid_price` returns `priceDirection: "quote_per_base"`, so a `SUI_USDC` result is USDC per SUI.
+`read_get_deepbook_mid_price` returns `priceDirection: "quote_per_base"`, so a `SUI_USDC` result is USDC per SUI.
 
 The price is the pinned SDK `midPrice` result with `source.precision: "deepbook_v3_to_fixed_9_js_number"`.
 
 The response includes `priceSemantics` and `userAnswerUse.cannotAnswer` for unsupported conclusions. It is a DeepBook pool snapshot, not a global market price, fiat USD cash-out estimate, external market-price conversion, USDC/USD peg assumption, quote-vs-mid slippage calculation, effective quote price, price-impact calculation, venue comparison, best-route claim, route recommendation, transaction-building input, signing data, signing readiness, P&L, or cost basis.
 
-If the SDK returns a non-positive or non-finite mid price, both `read.get_deepbook_mid_price` and `read.inspect_deepbook_orderbook` fail closed with `quote_unavailable`.
+If the SDK returns a non-positive or non-finite mid price, both `read_get_deepbook_mid_price` and `read_inspect_deepbook_orderbook` fail closed with `quote_unavailable`.
 
-`read.get_deepbook_usdc_price_history` returns DeepBookV3 official Indexer USDC candle evidence for a requested official interval.
+`read_get_deepbook_usdc_price_history` returns DeepBookV3 official Indexer USDC candle evidence for a requested official interval.
 
 Inputs require `start` and `end` as canonical ISO 8601 UTC timestamps and exactly one selector: `poolName`, `assetSymbol`, or `coinType`. Optional `interval` accepts only official Indexer interval values and defaults to `15m`.
 
@@ -148,7 +148,7 @@ USDC in this tool is the token-denominated quote asset for the returned official
 
 This output is not a live quote, execution price, historical mid price, global market price, fiat USD cash-out estimate, external market-price conversion, USDC/USD peg assumption, route recommendation, best route, transaction-building input, signing data, signing readiness, P&L, tax evidence, cost basis, user-account transaction history, or user-account balance history.
 
-`read.get_deepbook_usdc_price_at_time` uses the same official Indexer candle source and selector rules, but accepts one `targetTime` instead of a range. Optional `interval` defaults to `15m`. Optional `maxDistanceMinutes` limits how far from the target time the tool may select a candle.
+`read_get_deepbook_usdc_price_at_time` uses the same official Indexer candle source and selector rules, but accepts one `targetTime` instead of a range. Optional `interval` defaults to `15m`. Optional `maxDistanceMinutes` limits how far from the target time the tool may select a candle.
 
 Successful at-time responses return:
 
@@ -162,21 +162,21 @@ Successful at-time responses return:
 
 `status: "no_price_in_search_window"` means no official Indexer candle was available inside the bounded search window. The tool does not synthesize a price, interpolate, carry forward a previous candle outside the search window, or perform an on-demand chain-history scan.
 
-Internal read cards are opened with `ui.open_account`, `ui.open_receipt`, and `ui.open_chart`. They require a client that provides MCP Apps. A client without that UI receives `ui_unavailable`; ordinary MCP read tools remain available.
+Internal read cards are opened with `ui_open_account`, `ui_open_receipt`, and `ui_open_chart`. They require a client that provides MCP Apps. A client without that UI receives `ui_unavailable`; ordinary MCP read tools remain available.
 
 | Tool | Input and result |
 | --- | --- |
-| `ui.open_account` | Optional explicit Sui address; otherwise use the selected read account only when its wallet connection is currently usable. Missing input returns `input_required` with no card or source query. Shows name, address, coin totals, Display NFTs, other-object count, fetched time and enumeration limits; no address form. |
-| `ui.open_receipt` | A transaction digest is required for a query. Omission returns `input_required` with no card or source query. The result view shows execution outcome, balance changes, net gas, sender, digest and fetched time; no hash form. No wallet or active account is needed. Normalized receipt evidence and private display metadata remain available under their existing boundaries. |
-| `ui.open_chart` | Optional `poolName`, official Indexer interval, UTC `startTimeMs`/`endTimeMs`, and candle `limit`. Shows one selected USDC pool as candlesticks and volume. Range shortcuts fill these query fields. The initial time axis uses the saved request boundaries, with price-free whitespace at boundaries without candles; omitted boundaries use the returned data. |
+| `ui_open_account` | Optional explicit Sui address; otherwise use the saved Sui read address only when its wallet connection is currently usable. Missing input returns `input_required` with no card or source query. Shows name, address, coin totals, Display NFTs, other-object count, fetched time and enumeration limits; no address form. |
+| `ui_open_receipt` | A transaction digest is required for a query. Omission returns `input_required` with no card or source query. The result view shows execution outcome, balance changes, net gas, sender, digest and fetched time; no hash form. No wallet or active account is needed. Normalized receipt evidence and private display metadata remain available under their existing boundaries. |
+| `ui_open_chart` | Optional `poolName`, official Indexer interval, UTC `startTimeMs`/`endTimeMs`, and candle `limit`. Shows one selected USDC pool as candlesticks and volume. Range shortcuts fill these query fields. The initial time axis uses the saved request boundaries, with price-free whitespace at boundaries without candles; omitted boundaries use the returned data. |
 
 A missing Account/Receipt target returns `data: { kind, status: "input_required", field, message }`, with no permission, card ID, saved resource or query. A host-rendered view shows the message without input controls or polling. Other errors remain errors.
 
-A successful card-creating response contains a `cardId`, `kind`, `state`, `revision`, timestamps, original `input`, and available `data`. `inputRemainingMs` is computed by the backend; `pollAfterMs` controls sequential observation of a running read. Card `ready` permits a read selection, not signing or payment readiness. An initial supplied input uses the same admission path as a form submission. DB states are `ready`, `running`, and `closed`, with terminal reasons `completed`, `expired`, `failed`, or `server_restarted`.
+A successful card-creating response contains a `cardId`, `kind`, `state`, `revision`, timestamps, original `input`, and available `data`. `inputRemainingMs` is computed by the backend; `pollAfterMs` controls sequential observation of a running read. Card `ready` permits a read selection, not signing or payment readiness. An initial supplied input uses the same admission path as a form submission. DB states are `ready`, `running`, and `closed`, with terminal reasons `completed`, `expired`, `failed`, `server_restarted`, or `cancelled`.
 
 When Chart pool choices cannot be prepared, the backend creates a `closed`/`failed` card with a readable error. No selection is admitted and no candles are requested. Reading that card again returns the stored failure; retrying the source requires an explicit request for a new card. A database access or write failure is a tool error, not a successfully stored failure card.
 
-`ui.read_card` and `ui.submit_card` are app-only operations. Permission travels in UI metadata and is absent from model content and public saved resources. Input submission carries card ID, permission, expected revision and typed input. There is no first-view ownership or open/close operation. Recreated frames and chat returns read the same DB record. Identical duplicate input returns the stored request; conflicting input returns `card_conflict` and the authenticated current snapshot in error details. Invalid input returns `invalid_card_input` without consuming a valid selection. Invalid permission exposes neither current state nor private display data.
+`ui_read_card` and `ui_submit_card` are app-only operations. Permission travels in UI metadata and is absent from model content and public saved resources. Input submission carries card ID, permission, expected revision and typed input. There is no first-view ownership or open/close operation. Recreated frames and chat returns read the same DB record. Identical duplicate input returns the stored request; conflicting input returns `card_conflict` and the authenticated current snapshot in error details. Invalid input returns `invalid_card_input` without consuming a valid selection. Invalid permission exposes neither current state nor private display data.
 
 Receipt model results retain input kinds/indices/object references and identify input values and PTB as UI-only details. Receipt and completed Review show the executed transaction graph above their result summary when bound private metadata is available. Technical records are in one Details slide. Missing metadata leaves the saved result available without inventing a graph. The typed `sui-mcp/receipt-display` metadata binds display data to the same card ID, transaction digest and revision. It carries individual on-chain Pure input values and the graph derived from those inputs; never a serialized transaction, signature or wallet credential. Public saved resources contain only the model projection. A missing UI detail payload means unavailable display data, not that a transaction had no inputs.
 
@@ -210,8 +210,8 @@ current instruction or proof that a cancellation succeeded.
 A failed Request delivery never triggers a repeated signature request. Same-card
 reads still follow the existing expiry hints; a confirmed unadmitted stale review
 may prepare new conditions, which require a new Request click. Failed initial
-preparation can be retried for the displayed current wallet/account after another
-state check. A delayed action response does not clear an intervening read failure:
+preparation can be retried for the Review's original wallet/account after another
+state check, while that exact connection remains usable. A delayed action response does not clear an intervening read failure:
 the card retains explicit state-read recovery until a current read succeeds.
 
 Overlapping successful responses for the same card may arrive out of order.
@@ -219,7 +219,7 @@ A lower revision does not replace displayed facts or create a read error, and
 required observations continue from the retained current state. Equal revisions
 are validated rather than discarded. Actual error responses remain failures.
 
-An unconfirmed Cancel review, wallet disconnection, account selection or stop
+An unconfirmed Cancel review, wallet disconnection or stop
 request retains its error and pauses new automatic preparation/pairing in that
 View. State reads still report already admitted work. Existing permitted actions
 can be retried explicitly. An unconfirmed Cancel review followed by expired
@@ -231,12 +231,12 @@ pause does not claim that the backend or other frames stopped; new frames use
 the actual saved state.
 
 Connect eligibility and its SQLite admission use the same current-owner
-connection set. An unexpired connected wallet, pending approval or pending disconnection
+connection and SDK observation. An unexpired connected wallet, pending approval or pending disconnection
 removes `connect` from `allowedActions` and prevents a second pairing even when
 different cards race with current revisions. The losing card gets the existing
-conflict response and its own snapshot, without another card's private QR or
-permission. Later availability requires an explicit Retry for that failed
-attempt; a saved-state read does not repeat it.
+fixed no-request result, without another card's private QR or permission. It ends
+without pairing; later availability requires a new prompted Connect card. A saved
+result read cannot reopen it.
 
 Failed attempts whose exact target or command authority has ended are shown as
 earlier attempts, without a command retry or a saved-read button solely for that
@@ -244,12 +244,14 @@ old failure. The current stored result remains authoritative; natural completion
 is not proof that an earlier stop was accepted. This includes failed automatic
 connection/preparation and an unadmitted Request when its card or review expires.
 Card action expiry does not end an already admitted request or result lookup.
-For unadmitted disconnection or account selection, the exact target in
-`connections` can establish that the earlier attempt has ended even while the
-card input remains available. An account returning later does not revive that
-attempt or clear its local automatic pause; the user can make a new selection.
-If an unknown result needs another lookup after the card's action period, request
-`session.open_review_management` with the same `reviewSessionId` and `attemptId`.
+For an unadmitted disconnection, compare the attempted connection ID with the
+current `connection` and confirmed SDK facts. A replacement connection cannot
+be the target of that earlier attempt. Missing or unavailable observations alone
+do not establish termination. A later state change does not revive the attempt
+or clear its local automatic pause. A dedicated Disconnect card remains limited
+to its original connection; a new purpose card is required for an operation on the current connection.
+Once a result card has ended, request another lookup when needed through
+`session_open_review_management` with the same `reviewSessionId` and `attemptId`.
 It adds no signing authority and does not repeat the transaction.
 
 A confirmation that joins an older in-flight read retains its obligation. After
@@ -258,25 +260,25 @@ an actual read failure instead requires explicit recovery. A valid current read
 also clears an identified card's earlier opening-projection error. Neither path
 repairs malformed permissions or changes the card's target.
 
-Review displays one current wallet candidate without a dropdown. Its current authenticated View executes the backend-projected `automaticAction` to prepare or renew stale verified conditions. Multiple valid connections block preparation and signing until the user resolves them with targeted Disconnect actions. A failed computation requires explicit retry. `review_evidence_stale` distinguishes invalidated verified material from a failed quote/computation. Only Request wallet approval authorizes a signature request for the displayed revision. Result cards show the correct PTB, actual chain facts and one Details slide; Review keeps earlier estimates in its Reviewed conditions section.
+Review displays its exact account and wallet connection without a dropdown. Its current authenticated View executes the backend-projected `automaticAction` to prepare or renew stale verified conditions. A new Review uses the one current fixed Sui account and connection. Its public input has no account/connection selector. Disconnecting requires a new Review for the replacement connection. A failed computation requires explicit retry. `review_evidence_stale` distinguishes invalidated verified material from a failed quote/computation. Only Request wallet approval authorizes a signature request for the displayed revision. Result cards show the correct PTB, actual chain facts and one Details slide; Review keeps earlier estimates in its Reviewed conditions section.
 
 Cards provide no clipboard actions or copy buttons. Their text remains readable and selectable. Chart values remain USDC-denominated source candles, not fiat USD, peg guarantees, route advice, portfolio valuation or P&L. Cards do not fetch chain/Indexer endpoints directly or poll completed results.
 
 
-`read.quote_deepbook_action` and `read.quote_deepbook_display_amount` use the pinned DeepBook transaction builder quote functions.
+`read_quote_deepbook_action` and `read_quote_deepbook_display_amount` use the pinned DeepBook transaction builder quote functions.
 
 Input kinds:
 
-- `read.quote_deepbook_action`: `quantitySemantics.inputAmountKind: "raw_u64"`.
-- `read.quote_deepbook_display_amount`: `inputAmountKind: "display_source_amount_converted_to_raw_u64"`.
+- `read_quote_deepbook_action`: `quantitySemantics.inputAmountKind: "raw_u64"`.
+- `read_quote_deepbook_display_amount`: `inputAmountKind: "display_source_amount_converted_to_raw_u64"`.
 
 Quote responses include `quantitySemantics` before the public quote values. The same response also includes `userAnswerUse.cannotAnswer` for unsupported conclusions such as payment coverage, shortfall contribution, route-dependent payment support, fiat USD cash-out, external market conversion or lookup, USDC/USD peg assumptions, P&L, cost basis, price impact, mid-price slippage, venue comparison, and route recommendation.
 
 `canUseForPaymentAnswer: false`, `canUseForShortfallAnswer: false`, `doNotCombineWithPaymentAnswer: true`, and `paymentAnswerUseBlockedReason` mean quote output is a price estimate only. Do not add it to the payment amount, coverage status, or shortfall amount in a user answer.
 
-`requiredPaymentAnswerTool: "read.preview_intent_evidence"` and `requiredPaymentAnswerField: "responseSummary"` identify the tool and field to use for payment amount and shortfall answers.
+`requiredPaymentAnswerTool: "read_preview_intent_evidence"` and `requiredPaymentAnswerField: "responseSummary"` identify the tool and field to use for payment amount and shortfall answers.
 
-The same response-local rule is also exposed as `userAnswerUse.followUp.tool: "read.preview_intent_evidence"` and `userAnswerUse.followUp.answerFields: ["responseSummary"]`.
+The same response-local rule is also exposed as `userAnswerUse.followUp.tool: "read_preview_intent_evidence"` and `userAnswerUse.followUp.answerFields: ["responseSummary"]`.
 
 Raw quote source:
 
@@ -287,7 +289,7 @@ Raw quote source:
 
 Returned raw fields are quote evidence only. They are not final min-out values, effective prices, price-impact calculations, venue comparisons, best-route claims, transaction-building inputs, signing data, or signing readiness.
 
-`read.quote_deepbook_display_amount` accepts `amountDisplay` as the source coin input amount for the requested `direction`.
+`read_quote_deepbook_display_amount` accepts `amountDisplay` as the source coin input amount for the requested `direction`.
 
 It does not accept output target amounts or inverse quote requests.
 
@@ -310,7 +312,7 @@ This quote is not a settlement asset choice.
 
 ### DeepBook account inventory
 
-`read.summarize_deepbook_account_inventory` uses the selected account only when its current wallet connection is usable, to discover DeepBook BalanceManager addresses. It has no explicit public-address input. Without a usable default, a missing selection returns `active_account_not_set`, and a stored but ineligible selection returns `input_invalid`; both carry `details.reason: "connected_account_required"` and `details.followUp` naming `session.get_interaction_status` and its connection/availability/pending/default-account fields. These responses do not request an address. Unavailable wallet operations return `wallet_unavailable` with the existing safe reason and recovery message.
+`read_summarize_deepbook_account_inventory` uses the saved Sui read address only when its current wallet connection is usable, to discover DeepBook BalanceManager addresses. It has no explicit public-address input. A missing or cleared read address returns `active_account_not_set`; a stored but ineligible address returns `input_invalid`. Both carry `details.reason: "connected_account_required"` and `details.followUp` naming `session_get_interaction_status` and its wallet availability, current connection, pending operations and `assetReadAccount`. A cleared read address is reported separately from disconnection and cannot be bypassed with a typed address. These responses do not request an address. Unavailable wallet operations return `wallet_unavailable` with the existing safe reason and recovery message.
 
 With both `poolKey` and `managerAddress`, it checks pool account existence and returns display-like account inventory:
 
@@ -332,9 +334,9 @@ Returned `quantitySemantics.kind: "deepbook_display_number"` means these `number
 
 They are not raw balances, route liquidity, funding sources, withdrawal readiness, transaction-building inputs, signing data, or signing readiness.
 
-`read.summarize_wallet_assets` accepts an optional `account` input for public-address coin balance snapshots.
+`read_summarize_wallet_assets` accepts an optional `account` input for public-address coin balance snapshots.
 
-When `account` is omitted, it uses the same connection-qualified default as `session.get_interaction_status.assetReadAccount`. No stored selection returns `active_account_not_set` with `details.action: "provide_account"`; a stored but ineligible selection returns `input_invalid` with `details.reason: "address_required"`. Both request an explicit address. SDK unavailability is not reported as absence of a stored account.
+When `account` is omitted, it uses the same connection-qualified read address as `session_get_interaction_status.assetReadAccount`. A missing or cleared read address returns `active_account_not_set` with `details.action: "provide_account"`; a stored but ineligible address returns `input_invalid` with `details.reason: "address_required"`. Both request an explicit address. SDK unavailability is not reported as absence of a stored account.
 
 An explicit `account` read does not prove ownership, store the address as a known wallet, or create active account context.
 
@@ -356,7 +358,7 @@ If Sui metadata is unavailable, DeepBook-registered tokens can use pinned scalar
 
 If no verified decimals exist, `unit.status` is `unavailable` and clients must not infer decimals from the token symbol.
 
-`read.classify_wallet_assets` reuses the same explicit-address or active-account balance, unit, display, metadata cache behavior, and wallet snapshot quantity semantics as `read.summarize_wallet_assets`.
+`read_classify_wallet_assets` reuses the same explicit-address or active-account balance, unit, display, metadata cache behavior, and wallet snapshot quantity semantics as `read_summarize_wallet_assets`.
 
 It wraps each coin balance in `classification.assetClass: "coin_balance"`, `classification.balanceStatus`, and role labels such as `gas_candidate` or `deepbook_registered`.
 
@@ -372,7 +374,7 @@ It also would not make them route liquidity, payment readiness, portfolio comple
 
 The classifier does not create funding plans, choose routes, create review sessions, return transaction bytes, or produce signing material.
 
-`read.list_settlement_asset_groups` returns the supported natural-language settlement asset groups.
+`read_list_settlement_asset_groups` returns the supported natural-language settlement asset groups.
 
 In this release, `SUI_USD_SETTLEMENT_ASSETS` maps aliases such as dollar, dollars, USD, USD-like, stablecoin, stablecoins, and Korean dollar-word wording.
 
@@ -382,7 +384,7 @@ The asset group output includes included assets, excluded assets, source authori
 
 This is static SDK registry evidence, not live liquidity, fiat USD support, payment execution, route recommendation, or signing readiness.
 
-`read.summarize_settlement_asset_group_parity` returns internal parity evidence for a supported settlement asset group.
+`read_summarize_settlement_asset_group_parity` returns internal parity evidence for a supported settlement asset group.
 
 In this release it measures USD-denominated group assets against a declared reference asset. The default reference is USDC, but only as `measurement_reference_not_settlement_choice`.
 
@@ -398,19 +400,19 @@ Use `responseSummary` for questions about internal USD-denominated parity, highe
 
 This tool is not settlement-token choice, fiat USD cash-out, external market lookup, USDC/USD peg assumption, payment readiness, best route, route recommendation, transaction building, signing readiness, P&L, or cost basis.
 
-`read.preview_intent_evidence` accepts natural-language settlement evidence intents.
+`read_preview_intent_evidence` accepts natural-language settlement evidence intents.
 
 Use it this way:
 
 - Coverage question: `intentKind: "cover_payment_like_amount"`, `denomination: "dollar"`, and `requiredDisplayAmount: "1000"`.
 - AssetGroup-total question: `intentKind: "summarize_settlement_asset_group_balance"` and `denomination: "dollar"` without a target amount.
-- Account scope: explicit public `account` or a selected account with a usable current wallet connection.
+- Account scope: explicit public `account` or the saved Sui read address with a usable current wallet connection.
 
 For settlement-asset-only coverage, shortfall, and balance-total answers, use `responseSummary`.
 
 The response also exposes `userAnswerUse.answerFields` with the response-specific answer path.
 
-`responseSummary.answerCompleteness.answerCompleteFor` names the answer class. `responseSummary.answerCompleteness.requiredAnswerFields` names the fields required for that class. Do not call quote tools for the same question when `responseSummary.doNotCallQuoteToolsForThisQuestion` is `true`. If `toolAvailability.requiredToolsAvailable` is also `true`, answer from `responseSummary` and do not call `read.classify_wallet_assets`, `read.summarize_wallet_assets`, or quote tools to look for other source tokens for that same coverage, balance-total, or shortfall question. If quote tools were already called, do not use quote output for the payment amount, coverage status, or shortfall.
+`responseSummary.answerCompleteness.answerCompleteFor` names the answer class. `responseSummary.answerCompleteness.requiredAnswerFields` names the fields required for that class. Do not call quote tools for the same question when `responseSummary.doNotCallQuoteToolsForThisQuestion` is `true`. If `toolAvailability.requiredToolsAvailable` is also `true`, answer from `responseSummary` and do not call `read_classify_wallet_assets`, `read_summarize_wallet_assets`, or quote tools to look for other source tokens for that same coverage, balance-total, or shortfall question. If quote tools were already called, do not use quote output for the payment amount, coverage status, or shortfall.
 
 `responseSummary` exposes:
 
@@ -452,7 +454,7 @@ Review activity tools read only local Sui MCP review evidence.
 
 They are not complete wallet transaction history, gas history, P&L, or external wallet activity.
 
-The optional `account` input is a read filter and does not change the active account context. If `account` is omitted, the tools use the active account context.
+The optional `account` input is a read filter and does not change the saved read address. If `account` is omitted, the tools use stored read context without requiring a live wallet. If that context is absent or cleared, they return `active_account_not_set` with `details.action: "provide_account"`. Supply an explicit Sui address for the read; this error does not establish that the wallet is disconnected.
 
 Successful review activity responses include `dataScope`, `accountSource`, `lowSampleWarning`, and `lowSampleThreshold`.
 
@@ -460,7 +462,7 @@ The current `lowSampleThreshold` is 5 local records, from `REVIEW_ACTIVITY_LOW_S
 
 When `lowSampleWarning` is true, treat counts as sparse local evidence and avoid drawing behavior patterns.
 
-For `read.list_review_activity`, `dataScope.recordCount` is the full matching local review count.
+For `read_list_review_activity`, `dataScope.recordCount` is the full matching local review count.
 
 The returned `activities` array can be shorter when `truncated.activities: true`.
 
@@ -468,25 +470,25 @@ The returned `activities` array can be shorter when `truncated.activities: true`
 
 MCP error responses intentionally omit `structuredContent`; clients should read the JSON error payload from `content[0].text`.
 
-`read.get_review_session_detail` transition rows include `isNoOp`.
+`read_get_review_session_detail` transition rows include `isNoOp`.
 
 A no-op transition is an observed lifecycle call whose stored status did not change, such as a repeated open or reconnect event.
 
-`read.get_review_session_detail.userAnswerUse.answerFields` lists `execution` only when the response includes an `execution` object. If `execution` is absent and `userAnswerUse.cannotAnswer` includes `stored_review_execution_result_without_execution_field`, the response cannot answer a stored execution-result question.
+`read_get_review_session_detail.userAnswerUse.answerFields` lists `execution` only when the response includes an `execution` object. If `execution` is absent and `userAnswerUse.cannotAnswer` includes `stored_review_execution_result_without_execution_field`, the response cannot answer a stored execution-result question.
 
 Funnel summaries count distinct reviews and do not treat repeated no-op rows as additional completed steps.
 
 Sui activity tools are user-requested read surfaces, not a background indexer.
 
-Summary activity tools omit full transaction details by design. Digest-level detail is exposed by `read.inspect_sui_transaction`.
+Summary activity tools omit full transaction details by design. Digest-level detail is exposed by `read_inspect_sui_transaction`.
 
 Tool roles:
 
-- `read.inspect_sui_transaction` performs a single digest lookup.
-- `read.scan_sui_account_activity` requests up to 100 account transactions for `affected` or `sent` relationship.
-- `read.summarize_sui_activity_scan` reuses the live scan path and adds deterministic `analysis` without returning full `details`.
-- `read.scan_sui_function_activity` and `read.summarize_sui_function_activity_scan` return transactions the selected account sent that called one full `package::module::function`.
-- `read.summarize_sui_account_activity` reads only local SQLite stored facts.
+- `read_inspect_sui_transaction` performs a single digest lookup.
+- `read_scan_sui_account_activity` requests up to 100 account transactions for `affected` or `sent` relationship.
+- `read_summarize_sui_activity_scan` reuses the live scan path and adds deterministic `analysis` without returning full `details`.
+- `read_scan_sui_function_activity` and `read_summarize_sui_function_activity_scan` return transactions the selected account sent that called one full `package::module::function`.
+- `read_summarize_sui_account_activity` reads only local SQLite stored facts.
 
 This is a bounded provider page, not complete wallet history.
 
@@ -512,7 +514,7 @@ Stored summary boundaries:
 - Stored summaries aggregate local account-level facts from digest lookups, account scans, or sent-function scans.
 - Scan kind is internal provenance.
 - Stored summary input does not accept `kind`, `function`, or function-history filters.
-- Legacy `summary` is the backward-compatible shallow count block.
+- `summary` contains the stored transaction counts.
 - `analysis.overview.transactionCount` mirrors the stored count.
 - `analysis.overview.analyzedTransactionCount` reports how many returned rows fed the richer aggregations.
 - `transactionDetailAvailability` counts returned stored rows with `details`. `userAnswerUse.answerFields` includes `transactions[].compact` and `transactions[].details` only when `transactionDetailAvailability.allReturnedTransactionsHaveDetails: true`.
@@ -529,8 +531,9 @@ Coverage and pagination boundaries:
 
 Local persistence boundaries:
 
-- The default account is the active account.
-- An explicit account is allowed as a read filter.
+- An omitted account uses the saved read address without a live-wallet requirement.
+- A missing or cleared address returns `active_account_not_set` with `details.action: "provide_account"`; it does not establish wallet disconnection.
+- An explicit account is allowed as a read filter and does not restore implicit context.
 - The local DB stores only normalized facts for transactions tied to a known account by returned sender or balance-change owner facts.
 - Function activity scans store only sender-matching rows for known accounts.
 - Provider-returned account scan rows that cannot prove the local relation can appear in the current response but are counted as skipped for storage.
@@ -653,7 +656,7 @@ The local sender check is defensive. A non-conforming provider row is dropped fr
 
 For a stored known-account scan, the dropped row contributes only to skipped counts. It is not treated as valid function history.
 
-`read.scan_sui_account_activity` and `read.scan_sui_function_activity` return pagination and coverage fields:
+`read_scan_sui_account_activity` and `read_scan_sui_function_activity` return pagination and coverage fields:
 
 - `hasMore`;
 - `continuationCursor`;
@@ -671,22 +674,22 @@ Provider cursors are opaque and best-effort. Cursor rejection returns a safe err
 
 Tool source comparison:
 
-- `read.scan_sui_account_activity`, `read.scan_sui_function_activity`, `read.summarize_sui_activity_scan`, and `read.summarize_sui_function_activity_scan` return live GraphQL row facts.
+- `read_scan_sui_account_activity`, `read_scan_sui_function_activity`, `read_summarize_sui_activity_scan`, and `read_summarize_sui_function_activity_scan` return live GraphQL row facts.
 - Those live rows include requested-account fields, transaction context, and `detailLookup` references but no full `details`.
 - The summary tools also return deterministic `analysis`.
-- `read.inspect_sui_transaction` is the full normalized detail path for a specific digest.
-- `read.summarize_sui_account_activity` reads only stored normalized facts from local SQLite.
-- `read.get_account_asset_timeline` reads only stored normalized facts from local SQLite, then can attach DeepBookV3 official Indexer USDC candle references for supported USDC-quoted assets.
+- `read_inspect_sui_transaction` is the full normalized detail path for a specific digest.
+- `read_summarize_sui_account_activity` reads only stored normalized facts from local SQLite.
+- `read_get_account_asset_timeline` reads only stored normalized facts from local SQLite, then can attach DeepBookV3 official Indexer USDC candle references for supported USDC-quoted assets.
 
-If stored or summary details are missing or capped, use the digest metadata with `read.inspect_sui_transaction` instead of inferring missing calls, balances, objects, events, gas, or errors.
+If stored or summary details are missing or capped, use the digest metadata with `read_inspect_sui_transaction` instead of inferring missing calls, balances, objects, events, gas, or errors.
 
 ### Stored Account Asset Timeline
 
-`read.get_account_asset_timeline` returns a stored local account asset-flow timeline for one account and UTC range.
+`read_get_account_asset_timeline` returns a stored local account asset-flow timeline for one account and UTC range.
 
 Inputs:
 
-- `account` is optional. If omitted, the tool uses active account context. Supplying an explicit public account does not create active account context or prove ownership.
+- `account` is optional. If omitted, the tool uses saved Sui read context with `accountSource: "active_account_context"`, without requiring a live wallet. Missing or cleared context returns `active_account_not_set` with `details.action: "provide_account"`; supply an explicit Sui address for this read. That error does not establish wallet disconnection. Supplying an explicit public account does not create saved read context or prove ownership.
 - `start` and `end` are ISO 8601 UTC timestamps. The requested range is half-open: `start` is included and `end` is excluded.
 - `interval` is optional and uses the same official Indexer interval values as DeepBook USDC price tools. Omitted `interval` uses `15m`.
 
@@ -694,7 +697,7 @@ Output fields:
 
 - `coverage` describes whether stored scan rows prove the requested range is complete, partial, or absent.
 - `status: "account_not_known"` means the explicit account is not a known local wallet in the local activity store. In that status the tool does not return `scanNeeded`.
-- `scanNeeded`, when present, names `read.scan_sui_account_activity` as the bounded scan tool that can gather local evidence for a known account.
+- `scanNeeded`, when present, names `read_scan_sui_account_activity` as the bounded scan tool that can gather local evidence for a known account.
 - `netFlowBars` are observed account-scoped raw token balance-change bars from stored normalized activity rows.
 - `balanceStatus` is currently `unavailable_no_balance_anchor`, and `balanceBars` is empty. Do not present net-flow bars as held balances.
 - `sourceTransactions` reports how many stored rows were read, returned, truncated, and detail-covered.
@@ -708,19 +711,22 @@ This tool does not run scans, start a background indexer, create a price cache,
 
 | Tool | Status | Purpose |
 | --- | --- | --- |
-| `action.prepare_sui_action_review` | Signable internal Review card | Creates a local review session and internal Review card for a supported swap action proposal. Account-bound DeepBook reviews may build local unsigned transaction material inside the review server, internally bind a Sui transaction digest to it, and derive object ownership, quote/policy, human-readable review, review-time simulation, and PTB visualization evidence. When every required evidence stage completes, an explicit Review card selection can admit backend-mediated WalletConnect approval for that exact transaction. The MCP tool does not return transaction bytes, signing data, or signing readiness. |
-| `action.prepare_external_proposal_review` | Non-signable review | Creates a local review session and internal Review card from an untrusted structured external proposal. It does not return transaction bytes. |
-
-`action.prepare_sui_action_review` is account-bound: a swap review computes
-its evidence (balances, transaction material, digest, simulation) for a
-specific sender, so the tool requires an active wallet account. Connect first
-with `session.create_wallet_connection`; with no active account the tool returns
-`active_account_not_set` with `details.action: "connect_wallet_connection"`
-instead of creating a proposal that can never be computed or signed. The
-Review card reads that account from the server as the single source of truth
-and never connects a wallet itself.
-
-`action.prepare_sui_action_review` accepts a protocol-neutral swap `intent`
+| `action_start_review` | Internal swap Review card | Creates a local review session and internal Review card for a supported swap action proposal. Account-bound DeepBook reviews may build local unsigned transaction material inside the review server, internally bind a Sui transaction digest to it, and derive object ownership, quote/policy, human-readable review, review-time simulation, and PTB visualization evidence. When every required evidence stage completes, an explicit Review card selection can admit backend-mediated WalletConnect approval for that exact transaction. The MCP tool does not return transaction bytes, signing data, or signing readiness. |
+| `action_create_proposal_review` | Non-signable review | Creates a local review session and internal Review card from an untrusted structured external proposal. It does not return transaction bytes. |
+
+`action_start_review` accepts only `intent` and verifies the requested adapter
+before resolving the current connection's fixed Sui address. Public `account` or
+`connectionId` selectors are not accepted. No usable current target returns
+`active_account_not_set` with `details.reason: "connection_required"`. An unexpected
+retained SDK session returns `input_invalid` with
+`details.reason: "connection_state_unconfirmed"`; inspect connection status and
+remove the unwanted session in the wallet app. An unavailable service returns
+`wallet_unavailable`. These errors create no empty Review or wallet effect.
+The accepted account and connection are fixed in the initial card, before
+preparation. Preparation, display and retry retain that same target, including
+when it becomes unavailable or its saved read address is cleared.
+
+`action_start_review` accepts a protocol-neutral swap `intent`
 (`type: "swap"`, `from.symbol`, `from.amount`, `to.symbol`,
 `maxSlippageBps`, optional `protocol`). The optional `protocol` field carries
 the protocol slug from the adapter registry (the same slug vocabulary as the
@@ -731,7 +737,7 @@ omitted; once several protocols support the same action the tool returns
 `intent.protocol` set - the server never picks a venue silently. An unknown
 slug returns `reason: "unknown_protocol"` with the available slugs.
 
-`action.prepare_external_proposal_review` accepts `proposal` with `type:
+`action_create_proposal_review` accepts `proposal` with `type:
 "payment"` or `type: "sui_action"`.
 
 Common required fields are `id`, `source`, `network: "sui:mainnet"`,
@@ -792,7 +798,7 @@ cards display those facts; neither the model nor the card receives bytes or
 signatures. Ordinary MCP calls cannot authorize signing or submission.
 
 After the wallet account is bound to a supported swap review session, the
-Review card and `session.get_review_status` may show protocol-specific quote
+Review card and `session_get_review_status` may show protocol-specific quote
 evidence and review-state checks for:
 
 - resolved DeepBook direct pool;
@@ -819,8 +825,9 @@ Completed stages are review progress only. If
 `transaction_material_build_or_verify` is completed, it means the local review
 server built unsigned transaction material and kept the bytes internal. If
 `digest_commitment` is completed, it means the server internally derived a Sui
-transaction digest from that stored material; the digest value and transaction
-bytes are not MCP or review-app outputs. Missing stages explain why the review
+transaction digest from that stored material; the schema-validated transaction-review contract may expose its commitment
+hash, while transaction bytes remain private. The hash is not signing authority
+or a wallet handoff. Missing stages explain why the review
 remains blocked. This lifecycle covers review evidence producer stages through
 review-time simulation. Wallet requests, signature verification, submission and
 chain observation are platform responsibilities after review, separate from
@@ -943,28 +950,95 @@ It is not a signing readiness signal. Review-state checks are pre-signing review
 
 | Tool | Input and result |
 | --- | --- |
-| `session.create_wallet_connection` | Opens an internal card with `intent: connect`, `disconnect` or `manage` (default). A connect-intent View automatically starts pairing after authenticated state confirmation; manage pairs only after the user chooses Connect wallet. A disconnect-intent card opens confirmation for a unique connected target. Disconnection requires target-specific confirmation and does not revoke onchain permissions. |
-| `session.get_wallet_connection` | Reads one `cardId`; public response excludes pairing and permission. |
-| `session.wait_wallet_connection` | Waits on one `cardId`, with an optional timeout up to 55 seconds; stops when user input is needed or the admitted connection/disconnection operation ends. |
-| `session.open_review_management` | Opens management for exact `reviewSessionId` + `attemptId`; no refresh or signing authority. |
-| `session.get_interaction_status` | Stored `activeAccount`, current-owner `connections`, `assetReadAccount`, wallet availability, and bounded pending connection/review lists. `assetReadAccount` is `available` with an account only when that selected address has a usable current connection; otherwise `address_required`. Stored `activeAccount` does not prove a live connection. `address_required` reports that no implicit default is available; it does not add an address input to connected-account-only tools. When wallet availability is unavailable, connection rows are last recorded facts, not proof of presence or absence. |
-| `session.get_review_status` | Current preparation status, review revision, optional current request and chain execution. |
-| `session.get_execution_result` | Observes the current explicitly referenced attempt and may read its known digest; never submits again. |
-| `session.wait_execution_result` | Bounded wait on the same request, separate from its signing and chain-observation deadlines. |
-
-`pendingWalletConnections.items[].status` is `input_required`,
-`awaiting_approval`, or `disconnect_pending`. The last value describes an
-admitted, running disconnect card, not a new wallet connection state. In card
-workflow data, a connection's optional `pendingAction: "disconnect"` is derived
-from that stored card operation. Its `status` remains the last confirmed
-connection fact. Such a connection is unavailable for a new account selection,
-review or signature request. `observe` remains true while disconnection can be
-observed, and becomes false on completion or unavailable wallet progress.
-While its SDK callback remains pending, unavailable progress preserves the
-operation and its last recorded connection state. Repeating the same selection
-does not send another SDK request.
-
-`session.get_interaction_status` reconciles stored connections against current SDK sessions before reading dependent review/request facts. This can record a connection change and invalidate a review or interrupt an unsubmitted request under the existing rules. It preserves the stored account selection and submitted chain results; it does not start pairing, SDK disconnect, signing or submission.
+| `session_open_wallet_connection_card` | Opens one Connect card with empty input. Authenticated View confirmation admits its one pairing. A current session or unsettled SDK work prevents another pairing. |
+| `session_open_wallet_disconnection_card` | Opens one Disconnect card bound to the exact current manageable session. The user confirms removal or cancels this card. It never starts pairing or targets a replacement session. |
+| `session_get_wallet_connection_card` | Reads one `cardId`, returning `{card, operation, walletAvailability, evaluatedAt}` without UI authority or QR data. `card` is the stored snapshot; `operation` is that card's exact current public journal entry or null before admission. |
+| `session_wait_wallet_connection_card` | Uses `cardId` and an optional timeout up to 55 seconds. Returns the same envelope plus `waitOutcome`, when that exact operation and SDK settlement end, no operation is admitted, service is unavailable, or the bounded wait ends. It accepts no operation selector. |
+| `session_open_review_management` | Opens management for exact `reviewSessionId` + `attemptId`; no refresh or signing authority. |
+| `session_get_interaction_status` | Stored `activeAccount`, current `connection`, optional `pendingOperation`, `assetReadAccount`, wallet availability, and bounded pending connection/review lists. `assetReadAccount` is `available` with an account only when the saved Sui read address has a usable current connection; otherwise `address_required`. Stored `activeAccount` does not prove a live connection. `address_required` reports that no implicit default is available; it does not add an address input to connected-account-only tools. When wallet availability is unavailable, connection rows are last recorded facts, not proof of presence or absence. |
+| `session_get_review_status` | Current preparation status, review revision, optional current request and chain execution. |
+| `session_get_execution_result` | Observes the current explicitly referenced attempt and may read its known digest; never submits again. |
+| `session_wait_execution_result` | Bounded wait on the same request, separate from its signing and chain-observation deadlines. |
+
+`pendingWalletConnections.items[]` identifies each admitted `operationId`, its
+card and connection. Status is `awaiting_approval`, `disconnect_pending` or
+`cleanup_pending`. Unadmitted cards are not pending effects. These statuses
+describe management work, not a new wallet connection state. Interaction status has an optional top-level `pendingOperation` describing the current connection's
+outstanding operation. It contains `operationId`, `action`
+(`connect` or `disconnect`), `status` (`running`, `completed`, `failed` or
+`stopped`) and `settlementPending`. It is derived from that connection's existing
+operation when it is running or still has pending SDK work; it is not another
+operation or authorization. Connection `status` remains the last confirmed
+connection fact. Pending work makes that connection unavailable for a new review
+or signature request. `observe` follows the operation's remaining progress and
+wallet availability. SDK settlement remains observable through get/wait independently of the immutable
+terminal card result. Unavailable progress preserves the operation and its last
+recorded connection state. Repeating the same selection does not send another
+SDK request.
+
+Match `pendingOperation.operationId` to the one operation fixed to that card,
+rather than inferring ownership from the action name. Closed card data is a
+recorded snapshot; the get/wait envelope's `operation` reports current settlement.
+A running disconnect is disconnecting; a completed disconnect is disconnected
+with any outstanding cleanup reported separately. It is not another connection
+request, and pending cleanup does not turn completion into failure or reopen a
+cancelled card.
+
+`connection.sessionPresence` reports whether the current connection has an SDK
+session: `present`, `absent`, `unusable`, or `unconfirmed`. Absence includes an
+exact current-session absence observation or a settled attempt without an approved
+session or recorded session reference. The latter requires confirmed current-run
+ownership, no pending connection work and no unrecorded SDK session. A missing
+observation for a recorded session is not absence. Unavailable service or an
+unconfirmed current-run observation is `unconfirmed`, never evidence of absence.
+This field does not change the saved connection status or grant input authority.
+A failed or expired connection may still need explicit disconnection. Once
+absence is confirmed and cleanup ends, the user can request a new Connect card.
+An ended card never offers a new connection. An existing Review bound to the ended connection requires a new Review
+after reconnection. A declined connection alone does not require a transaction
+Review. Stored connection messages in Details are historical facts, not current
+recovery instructions.
+
+Live card data is purpose-specific. Connect includes fixed `requestedChains`,
+its optional `operation` and permitted initial `automaticAction`. Disconnect
+includes its exact `connection` confirmation and optional `operation`. Review
+includes `review`, `reviewTarget`, optional exact `request`/receipt facts, and
+permitted preparation action. All expose `allowedActions`, progress, service
+availability and `nextStep` (`kind`, optional safe `message`/`reason`). This is
+advisory guidance, never authorization. Read-address overview and current
+`pendingOperation` belong to interaction status, not the live card payloads.
+
+| `nextStep.kind` | Meaning |
+| --- | --- |
+| `none` | No additional business recovery step is required by this evaluation. |
+| `wait` | Service startup or existing wallet/account work must finish before continuing. |
+| `check_status` | Read the indicated existing request or session; no new wallet effect is authorized. |
+| `open_settings` | Open the named Settings section for configuration or explicit service recovery. |
+| `request_disconnect` | Ask for a new exact-target Disconnect card before changing the connection. |
+| `request_review` | Ask for a new transaction Review; the old target cannot be replaced. |
+
+The backend's message identifies the applicable condition. A saved preparation
+error does not override this guidance. Reasons such as `operation_busy`,
+`connection_state_unconfirmed`, `connection_target_changed`,
+`wallet_networks_required` and `no_session_to_disconnect` describe the particular
+refusal; they grant no input or effect authority. Service `initializing` and
+`recovering` mean wait, not another restart request.
+
+A closed wallet card has no operation copy. Optional `result.pendingWork` records
+`approval_response`, `session_removal` or `session_status` at closure. Normal
+success does not create a recovery obligation from callback bookkeeping.
+Exact-operation get/wait still reports current `settlementPending`, including
+failed bookkeeping writes, without changing the card or replaying effects.
+A Connect card that sends no request has no network approval result or wallet
+name copied from another connection.
+
+Every Connect and Disconnect terminal snapshot is address-free from the moment
+it is recorded: no address, account, address-bearing chain states or QR payload
+is included. Its result, recorded time, wallet name and network approval statuses
+stay fixed. Current `connection.chainResults` belong to interaction status;
+stored history and transaction targets keep their exact addresses.
+
+`session_get_interaction_status` reconciles stored connections against current SDK sessions before reading dependent review/request facts. This can record a connection change and invalidate a review or interrupt an unsubmitted request under the existing rules. It preserves fixed addresses, read-address suppression and submitted chain results; it does not start pairing, SDK disconnect, signing or submission.
 
 When a disconnect callback ended but its completion write failed, current reads
 reconcile that stored operation without another SDK disconnect. If the SDK is
@@ -998,8 +1072,9 @@ neither a completed result nor a wait timeout. A command already committed
 before its response fails remains recorded and is recovered by reading it.
 
 Wallet startup and restoration are asynchronous dependencies of the parent
-service. Connect/manage cards remain available during initialization or failure
-so the user can inspect state. They do not provide service-restart input. Ordinary evidence
+service. Connect can explain service initialization or failure; Disconnect
+creation instead returns its applicable availability error without issuing a
+card. Neither card provides service-restart input. Ordinary evidence
 and saved-result reads do not await the SDK. The package supplies its project
 identifier; these failures do not ask the user for a project ID. SDK error bodies
 and private configuration values are not returned.
@@ -1012,19 +1087,16 @@ Review status is preparation state (`proposed`, `awaiting_wallet`,
 otherwise the preparation state. The `user_action_required` category and
 `pollingHint.userActionRequiredStatuses` identify states where the model-facing
 wait returns; they do not require a manual preparation click. A permitted live
-card can prepare automatically, while an ambiguous selection or failed update
-can require user input. Preparation is not an already pending wallet signature.
-
-A review's bound account is fixed; changing the active read account does not
-rebind it. Preparation choices and final admission use the same account rule.
-To update a review bound to A, select A again in a new Connect card. To review
-for B, request a new Review. An absent or incompatible read account removes the
-preparation choice without recording a preparation failure. The existing
-`review.error` card text distinguishes `Current account selection` guidance from
-a quoted saved review message. It does not decide permissions. Changing
-read context alone does not revoke an already valid A review's signing choice.
-Once a signing request consumes its card, retrying needs a new Review; a
-management card only refers to its existing exact attempt.
+card can prepare automatically, while a failed update
+can require an explicit retry. Preparation is not an already pending wallet signature.
+
+A Review keeps its exact account and connection. Clearing the saved read address does not change its target. Disconnecting means
+a replacement connection needs a new Review. A new target requires
+a new Review; another connection with the same address cannot take over.
+Preparation and signing still require that original binding's current grant,
+method/event scope, revision and material. Saved preparation errors remain
+separate from current eligibility. After signing admission, a management card
+only refers to its exact attempt and grants no new preparation or signature.
 
 Only `completed` has `executionResult`, with `attemptId`, review/plan IDs,
 `status: success | failure`, the exact digest, verified `chainReceipt` and time.
@@ -1032,7 +1104,10 @@ Only failed chain effects have `failureReason: chain_execution_failed`. Wallet
 rejection, local timeout, unavailable receipts and lost submission responses
 never fabricate a chain failure. An unknown result retains the digest for later
 reads. An explicit user stop after submission stops observation, not an on-chain
-transaction. Wallet connection changes or a requested disconnect stop pending
+transaction. Accepted `stop_waiting` also closes that card's business input;
+a new `session_open_review_management` card may explicitly inspect the same
+attempt later. A duplicate stop on the cancelled card does not stop work started
+from a newer card. Wallet connection changes or a requested disconnect stop pending
 submission permission with their own recorded reason; they do not claim that the
 user stopped waiting or that disconnection already succeeded. After submission,
 those connection events preserve independent observation of the same digest,
@@ -1043,7 +1118,7 @@ The original request's stored deadlines are unchanged by a management card.
 A new signature requires a newly reviewed revision and explicit selection.
 Identical duplicate selections return the admitted attempt; no response-loss
 path replays a financial request. Model tools may open/read cards; only scoped
-app-only `ui.act_card` performs a user selection. UI permission and pairing QR
+app-only `ui_act_card` performs a user selection. UI permission and pairing QR
 travel in private metadata. Bytes/signatures never reach the card or model.
 
 Connection observation uses 5 seconds and request observation 3 seconds.
@@ -1067,80 +1142,70 @@ for the last confirmed local SDK observation. Ordinary reads do not perform a
 fresh SDK check or prove that the wallet app is online. Implicit asset reads
 require an available service and a usable observation in that same run.
 
-Wallet-dependent app inputs (`connect`, `disconnect`, `use_account`,
-`prepare_review`, `request_signature`) carry the displayed `walletRunId`. A stale
-run cannot be transferred to a replacement SDK. Backend commands verify the
-session, account, chain, method and stored authority at their decision boundary.
-`acceptedWalletRunId` identifies a card's stored wallet input, when present; it
-is not new permission.
-
-Connect workflow data includes optional `connectionAction: "connect" | "disconnect"`
-only when that action was admitted on the card. The connection's current status
-is separate from the card's original operation. In particular, failed disconnect
-means remote removal was not confirmed, including after local service recovery.
-
-`connectionConflict`, when present, is
-`{ reason: "multiple_connections", connectionIds: [...] }`. It contains distinct
-IDs for the multiple live, known connections that prevent selecting a usable
-wallet. It appears in Connect/Review `data`, the connection wait's `card.data`,
-interaction status and relevant review-status responses. Its shared schema and
-backend evaluation own the meaning; the View does not count rows to grant authority.
-`walletAvailability: available` only confirms the SDK dependency, not the absence
-of this conflict. `connections` describes recorded targets, not signing choices.
-In conflict, `assetReadAccount` is `address_required`; explicit-address reads and
-saved transaction results retain their independent access rules.
-
-A valid Connect card can still offer exact-target `disconnect` during conflict.
-`connect`, `use_account`, `prepare_review` and `request_signature` remain blocked.
-Each disconnection uses the same scoped UI permission, current run and revision,
-atomic admission and saved result as an ordinary single-connection disconnect.
-No first/newest/default wallet is silently selected. When one valid connection
-remains and no connection operation is pending, normal account use can resume;
-with no recorded or SDK-present connection and no pending operation, pairing is
-available. A failed disconnect does not prove remote removal.
-A failed or stopped product connection is not restored merely because its SDK
-session remains saved.
-
-`connectionConflict` is optional. Its absence alone is not proof of a usable
-connection or permission, especially in a historical result that does not need
-current wallet eligibility. Filling optional context never adds global cleanup
-writes or SDK requests to independent saved-result reads. `allowedActions`
-identifies permitted action types; it is not a list of visible buttons or model
-authority. Only the exact current target, UI permission and final admission
-checks allow a user action.
-
-Current Connect/Review `data` also publishes `usableConnectionId` and
-`assetReadAccount`. The first identifies the single connection eligible for
-account use at the backend's evaluation time; it is absent during conflict,
-pending connection work, unavailable service, or when no unexpired connection
-qualifies. It does not authorize a signature. `assetReadAccount` uses the same
-`available`/`address_required` shape as interaction status. Availability requires
-the stored default's source connection ID to match this connection and its
-address to remain approved. Matching the address in a different connection is
-insufficient. A source-less stored selection requires explicit Use address.
-
-`manageableConnectionIds` identifies exact recorded targets for a new scoped
-Disconnect, separately from account use. It includes nonpending connected records
-(including expired ones) and known terminal records whose current SDK session
-remains present. `connectionReconciliationRequired: true` identifies a present SDK
-session without a journal target after normal registration/cleanup ends. It
-blocks pairing, account use, preparation and signing while preserving SDK
-availability and saved outcomes. Remove the retained connection in the wallet
-app, then confirm absence through a complete current service observation.
-Settings/Troubleshooting can explicitly restart the service to inspect it again;
-restart alone does not remove or approve that connection. No private topic or synthetic target ID
-is exposed. Both fields also appear in interaction status; review status includes
-the reconciliation flag when current eligibility is needed.
-
-These fields are emitted together for current wallet eligibility. Independent
-historical or completed-result reads may omit them; omission does not establish
-current eligibility. Recorded `connections` and `activeAccount` remain facts,
-including expired records and a stored default that cannot currently be used.
-Views consume the published selection instead of counting recorded rows or
-using their own clock. An expired record may remain a disconnection target;
-expiry alone does not confirm remote removal. Final admission checks current
-backend facts again. Changing a read default never changes an existing review's
-account or grants new authority to an admitted transaction.
+Wallet-dependent app inputs (`connect`, `disconnect`, `prepare_review`,
+`request_signature`) carry the displayed `walletRunId`. A stale run cannot be
+transferred to a replacement SDK. Exact account, chain, method, grant revision
+and stored request authority are rechecked at admission. `acceptedWalletRunId`
+is a stored fact, not new permission.
+
+In `session_get_interaction_status`, `requestedChains` is the Settings scope for new connections. Singular `connection`
+publishes its frozen requested scope and `chainResults`: exactly one result per
+requested chain, either `bound` with an address, `not_approved`, `multiple_accounts`
+or `invalid_response`. One valid distinct address binds automatically. There is
+no account selector. Only that chain becomes unavailable on multiple/invalid data.
+Later observations cannot change those results. Current `grants`, `grantRevisions`
+and `chainStates` describe availability separately; methods remain account-local.
+Sui actions require the fixed Sui address and exact Sui method/event scope.
+Ethereum/Solana address fields do not provide financial authority.
+
+Both wallet gateways accept empty input. Disconnect creation checks the current
+exact session before issuing any card or permission. No recorded session or a
+confirmed absent settled session returns `input_invalid` with reason
+`no_session_to_disconnect`; unconfirmed or unrecorded presence uses
+`connection_state_unconfirmed`, and outstanding SDK work uses `operation_busy`.
+An unavailable service returns `wallet_unavailable` with its actual cause.
+These refusals create no card, operation or wallet effect. A present or unusable
+manageable session is fixed to the new Disconnect card; admission rechecks its
+run and impact. Changing wallet, address or approved network scope requires
+confirmed disconnection and a new explicit connection.
+
+Connect fixes requested networks and expected connection identity at creation.
+A confirmed change before admission closes it failed, sends no wallet request,
+and asks for a new card; restoring the earlier Settings value does not reopen it.
+Transient service initialization is not proof of such a change.
+
+`reviewTarget` retains the Review's original account and connection before,
+during and after preparation, even when that target is no longer usable.
+`allowedActions` lists currently permitted app input, computed for that exact
+target. A live Review's `nextStep` names the applicable recovery action and safe
+message; the view does not infer it from another connection's status. A replacement
+connection is never substituted for the original Review target.
+The interaction-status field `assetReadAccount` independently describes the fixed Sui read address as
+`available` with `account`, or `address_required`. Reason `read_address_cleared`
+means the user cleared the saved read address in Settings; this does not change
+fixed addresses or financial targets. `account_get_active_account` reports stored
+read context, not current eligibility. Restored public accounts never restore
+wallet or card authority.
+
+The status overview returns `connection: null` when no current connection is recorded.
+`canDisconnect` qualifies the current exact nonpending cleanup target, including
+terminal or older-generation records whose SDK session is present/unusable.
+Current-run confirmed absence and ended I/O permit a new explicit Connect; a
+historical failed/expired row alone is not absence evidence. Recovery can transfer
+cleanup ownership without restoring usage or old card/signing authority.
+
+In interaction status, `connectionReconciliationRequired: true` means current SDK presence has no approved
+product target after pending registration/cleanup. New pairing and financial
+inputs are blocked; saved evidence remains readable. Unrecorded sessions are not
+adopted by address or ordering. Remove unwanted retained sessions in the wallet
+app, then check again. Restarting inspects but neither approves nor removes them.
+Private topics are not exposed. An already admitted transaction still uses its
+original exact grant and submission checks.
+
+These are backend-evaluated facts, not View row counts or clock decisions.
+`allowedActions` is not model authority. Scoped UI permission and the exact target
+remain required at atomic admission. Reconnection cannot retarget a Review or an
+admitted transaction.
 
 Views additionally know whether their own confirmation read or command response
 is pending. They show neutral communication guidance without synthesizing DB
@@ -1189,7 +1254,9 @@ as uncertain without resending them. Neither event establishes cancellation or
 chain failure, and verified chain outcomes remain unchanged.
 
 Wallet-dependent commands rejected for dependency failure use the error kind
-`wallet_unavailable`, with a safe reason and message. Invalid input and domain
+`wallet_unavailable`, with the actual safe `walletAvailability` and message.
+An unavailable service includes its recorded `reason`; initializing/recovering
+states have no failure reason, so none is invented in error details. Invalid input and domain
 conflicts retain their own meaning. Validated app actions can also return a
 current snapshot with `error.code: "wallet_unavailable"`; invalid permissions
 and storage failures do not receive a success snapshot. Review creation can
@@ -1210,13 +1277,13 @@ retained when a later submission-admission check fails.
 
 ### Stored review activity
 
-`read.list_review_activity` accepts account/time/limit and independent ANDed
+`read_list_review_activity` accepts account/time/limit and independent ANDed
 `reviewStatus`, `requestStatus`, `executionStatus` filters. One row represents
 one review session and its explicit current attempt, including optional
 `currentAttemptId`, `reviewRevision`, `transactionDigest`, request and chain
 status. `dataScope.recordCount` counts the complete filtered scope before limit.
 
-`read.summarize_review_funnel` keeps the session denominator. `reviewStatusCounts`
+`read_summarize_review_funnel` keeps the session denominator. `reviewStatusCounts`
 is the preparation distribution; `requestStatusCounts` is a list of
 `{ requestStatus, count }` entries (including zero counts); `executionStatusCounts`
 has success/failure counts. Status strings are data values, not secret-like JSON
@@ -1237,7 +1304,7 @@ result counts as completed and as its verified chain outcome, but does not count
 as `everAwaitedChainResult` unless the session actually reached that stage in an
 attempt. This metric is neither a submission total nor a mandatory funnel step.
 
-`read.get_review_session_detail` returns the current `request`, full
+`read_get_review_session_detail` returns the current `request`, full
 `requestCount`, capped `requests`, revision-bound state snapshots and transitions
 with `domain` and optional attemptId. Each request may contain a verified
 `execution`; absent execution does not imply failure. Current request is returned
@@ -1250,29 +1317,66 @@ facts actually exist.
 
 | Tool | Status | Purpose |
 | --- | --- | --- |
-| `account.get_active_account` | Implemented | Reads the active wallet-account read context. |
-
-The user clears the default read address through a confirmed Settings card action. Ordinary MCP calls do not clear it.
+| `account_get_active_account` | Implemented | Reads the active wallet-account read context. |
+
+Settings provides Clear saved read address for the current Sui read context.
+The explicit app-only action `settings_clear_read_address` binds the displayed
+connection and address and records its effect atomically. It suppresses implicit
+read context for that connection across restarts, without disconnecting the wallet,
+changing fixed addresses or cancelling transaction requests. A fresh connection
+can supply its new fixed read address. Ordinary MCP calls cannot apply this effect.
+Changing from a restore upload to this effect invalidates the unused upload
+atomically with the saved result. File cleanup runs separately; a cleanup failure
+does not repeat the address-clearing effect or change its stored outcome.
 
 ## Settings Tools
 
 | Tool | Status | Purpose |
 | --- | --- | --- |
-| `settings.create_local_settings_session` | Implemented | Opens one internal Settings card with an unsubmitted draft. |
-| `settings.get_local_settings` | Implemented | Reads stored/effective Sui gRPC and GraphQL endpoints, sources and pending restart information. |
+| `settings_create_local_settings_session` | Implemented | Empty input opens Settings. `operationStatus: {operationId}` records a read-only status card. `backupDelivery: {operationId, fileId}` opens a new delivery-purpose card for the exact retained backup. These strict input forms cannot be mixed. |
+| `settings_get_local_settings` | Implemented | Reads requested wallet networks and stored/effective Sui endpoints, sources and pending restart information. |
 
-The gateway name is preserved. Its result is a `kind: "settings"` card snapshot
+The result is a `kind: "settings"` card snapshot
 and a public saved-result link. Card permission is only in `sui-mcp/card` UI
-metadata. There is no `settingsUrl`, `openTarget`, external page, HTTP mutation
+metadata. Every closed Settings card keeps its recorded result and has empty
+`allowedActions` and `allowedFileActions`. A fresh `operationStatus` card records
+current operation/cleanup facts once; its `createdAt` is that query's timestamp,
+and any stored operation result retains its original `recordedAt`. A current
+backup-delivery summary is recorded at the query time; do not infer a delivery
+completion time from that read timestamp. It issues no settings
+or file effect authority. A new `backupDelivery` card uses the original artifact,
+owner, epoch and expiry; creating it performs no transfer. Only its explicit Retry
+download or Save a local copy admits one attempt. A late exact Host report updates
+the domain journal only; it cannot change a closed card or initiate local saving. There is no `settingsUrl`, `openTarget`, external page, HTTP mutation
 API or compatibility alias. A client without internal cards receives
 `ui_unavailable`; ordinary model calls never apply a Settings change.
 
-Settings groups Address for viewing assets, Local records, Troubleshooting and Advanced
-settings. Confirmed changes are `clear_read_address`, `set_endpoint`,
-`restore_endpoint` (gRPC or GraphQL), `import_local_data`, `reset_local_data`
-and `restart_wallet_service`. Their preparation and confirmation use app-only
-`ui.act_card` with card permission, revision, operation identity and the exact
-preparation identity. Choosing or checking data does not apply it, contact a
+Settings groups Address for viewing assets, Wallet networks, Local records,
+Troubleshooting and Advanced settings. Address for viewing assets owns the
+`settings_clear_read_address` action described under Account Tools; clearing
+does not disconnect the wallet or change transaction targets.
+
+Wallet networks uses one explicit Save networks action:
+`settings_save_wallet_networks` with `operationId`, the selected `chains`, and
+`expectedChains` containing the saved scope displayed to the user. App-only
+`ui_act_card` checks the card permission, owner, data epoch, revision and expiry,
+compares the current saved scope with `expectedChains`, and reserves, saves and
+records the exact selection in one transaction. Its `data.draft.kind` is
+`set_wallet_networks`. There is no selection/preparation/confirmation
+sequence for this setting. An identical admitted save returns its stored result;
+a conflicting selection cannot replace it. Editing checkboxes does not save,
+pair a wallet or update an existing connection. Saving a changed scope atomically
+ends unsent Connect cards with different captured networks; restoring the earlier
+setting does not reopen those cards. Already admitted wallet requests retain their
+original scope.
+
+Endpoint changes (`set_endpoint`, `restore_endpoint` for gRPC or GraphQL),
+`import_local_data`, `reset_local_data` and `restart_wallet_service` retain a
+concrete impact confirmation. Selecting an endpoint change, reset or restart
+starts preparation directly; no intermediate Continue action is required.
+Selecting a backup file starts its checks. Final confirmation uses app-only
+`ui_act_card` with card permission, revision, operation identity and the exact
+preparation identity. Preparing or checking data does not apply it, contact a
 wallet or restart the SDK. Duplicate confirmation returns the same admitted
 result; MCP JSON-RPC request IDs do not authorize another operation.
 
@@ -1280,8 +1384,7 @@ Download backup uses one `settings_download_backup` action with `operationId`
 and an `attemptId` UUID, admitted atomically through the same scoped card input.
 It creates a `download_backup` operation and one immutable snapshot. It requires
 no separate selection, preparation or confirmation action. The same admitted
-start returns the same request; a different start cannot replace it. New
-`export_local_data` selection/preparation/confirmation is not accepted.
+start returns the same request; a different start cannot replace it.
 
 The card uses plain-language display labels without changing wire fields or
 enums. Wallet-service `available`, `initializing`, `recovering` and `unavailable`
@@ -1319,16 +1422,21 @@ failure-recovery policy.
 `file.sealed` records complete, integrity-checked transfer; it does not by itself
 establish a supported backup or permission to restore it. `settings_back` and user-data
 changes invalidate preparation while preserving a reusable complete file.
-The explicit `settings_prepare` action rechecks such a file; normal file
-selection requires no separate Preview action. A failed check preserves its
+The explicit `settings_prepare` action rechecks such a file; selecting a file
+starts its backup check without another action. A failed check preserves its
 current file/operation identity and correction path. Owner replacement ends old
 input; a new card must receive a new explicit file selection.
 
-A download's `result.completionScope: "backup_snapshot"` identifies the immutable
-preparation result, not a saved user file. `data.download` separately carries
+Physical snapshot readiness is intermediate: the delivery card stays running.
+Its final `result.completionScope: "backup_delivery"` records a Host report,
+verified local publication, cancellation, expiry or failure, not merely snapshot
+preparation. A Host acknowledgment is not proof that a user file was saved. `data.download` separately carries
 `availability` (`preparing`, `available`, `ended`), `currentAttemptId`, an optional
-`hostReport` (`acknowledged`, `not_completed`, `unconfirmed`, `unsupported`), `previousUnconfirmed`,
-`expiresAt`, `remainingMs`, `endReason` and `allowedFileActions`. End reasons are
+`hostReport` (`acknowledged`, `not_completed`, `unconfirmed`, `unsupported`, `not_sent`), `previousUnconfirmed`,
+`expiresAt`, `remainingMs`, `endReason` and `allowedFileActions`. `not_sent` is the
+View's report that transfer verification failed before any Host download request;
+it is not a Host rejection or non-support response. Like `unsupported`, it
+establishes no earlier Host copy for local-copy consent. End reasons are
 `acknowledged`, `cancelled`, `expired`, `server_restarted`, `data_replaced`,
 `source_unavailable`, `stored_locally` or `local_save_failed`. Replacement revokes the old card rather than preserving
 its input. The transfer period is thirty minutes from physical readiness;
@@ -1344,13 +1452,12 @@ absence was not established, including an unreported or not-completed request.
 
 A verified local `saved` result provides `data.backupLocation.directory` and
 `filename` for `<SUI_MCP_DATA_DIR>/backups/<operationId>.json`. Snapshot preparation
-and Host acknowledgment do not provide that local success. A recorded
-`export_local_data` result can also identify its existing file. These fields
-never identify the destination chosen by a Host download dialog. Existing user
+and Host acknowledgment do not provide that local success. These fields never
+identify the destination chosen by a Host download dialog. Existing user
 files are never deleted by private snapshot cleanup. Private publication
 identities and source inode/device facts are not public projection fields.
 
-Current `data.facts` uses the same connection selection and qualification rules
+Current `data.facts` uses the same fixed-address and connection qualification rules
 as interaction status, through a fact reader that does not settle another
 request's deadlines or callback writes. General counts describe stored records;
 reset/import preparation and confirmation independently settle deadlines and
@@ -1383,8 +1490,10 @@ recorded idempotently without resending bytes. Public fields include
 local operation, not a transaction's chain outcome. Optional
 `temporaryFileCleanup` separately reports `pending`, `failed` or `complete`;
 reading/checking it never repeats data replacement, backup creation, Host delivery or SDK
-replacement. Terminal business input remains closed; only the separate scoped
-file continuations in `data.download.allowedFileActions` may remain available. Public fields contain no raw file, transaction bytes, signatures,
+replacement. Terminal business input remains closed; the separate scoped
+file continuations in `data.download.allowedFileActions` are also closed. A new
+`backupDelivery` card is required for another attempt, while `operationStatus`
+reads current cleanup without granting a file effect. Public fields contain no raw file, transaction bytes, signatures,
 SDK credentials or card permission.
 
 A user-data writer invalidates its affected prepared confirmation atomically.
@@ -1403,8 +1512,8 @@ rejection can report a smaller raw chunk limit before any write. The backup chec
 strict current backup shape/references without endpoint contact. Confirmation
 verifies imported endpoints report Sui mainnet and rechecks current data and
 unsettled wallet/transaction work before atomic replacement.
-The prepared import includes `currentReadAddress` and `incomingReadAddress`
-(`null` when unset), the corresponding change and current/incoming record counts.
+The prepared import includes current/incoming record counts. Accounts and records
+are restored as history, never as active connection or signing authority.
 Both preparation and final admission settle request deadlines before capturing
 or comparing that decision basis. A newly unknown outcome requires a new backup check.
 Reset/import revoke other live permissions and preserve only their own read-only card/result.
@@ -1416,8 +1525,9 @@ automatic result checks. Stopping those checks does not shorten that window.
 They cannot cancel a transaction on Sui or disconnect a remote wallet session.
 The private connection-service journal is preserved; it is not imported from
 the public backup and cannot restore old permissions or signing authority.
-Reset clears the default address. A restored default requires its exact current
-connection ID and approved address. Outstanding connection SDK callbacks block
+Reset and restore invalidate the retained connection for financial use. Neither
+restores a live connection reference; the retained session remains a cleanup
+target until disconnected. A new explicit connection establishes new fixed addresses. Outstanding connection SDK callbacks block
 replacement along with unsettled transaction work.
 
 Endpoint preparation records the gRPC/GraphQL kind, current stored/effective
@@ -1429,18 +1539,19 @@ remains set; storing an endpoint does not modify that environment.
 
 Download creates and verifies one private snapshot with an exclusive generated
 file identity. Supported Host delivery does not publish a second permanent backup. Response or result
-storage failure never regenerates that snapshot. `ui.read_settings_file` is
+storage failure never regenerates that snapshot. `ui_read_settings_file` is
 app-only and binds card/permission, operation, file, offset, verified hash/length
 and the current `attemptId` for a private download. Its ordinary content carries
 range metadata; Base64 bytes appear only in `sui-mcp/settings-file` UI metadata.
 The View checks the complete hash before standard Host delivery.
 
-File continuations also use app-only `ui.act_card`:
+File continuations also use app-only `ui_act_card`:
 
 - `settings_retry_download`: `operationId`, `fileId`, `previousAttemptId` and a
-  new `attemptId`; atomically advances one attempt on the same live snapshot.
+  new `attemptId`; only a new backup-delivery card may admit this attempt on the
+  same retained snapshot. It consumes that card's one delivery input.
 - `settings_report_download`: `operationId`, `fileId`, `attemptId` and `status`
-  (`acknowledged`, `not_completed`, `unconfirmed`, `unsupported`); the same report is idempotent,
+  (`acknowledged`, `not_completed`, `unconfirmed`, `unsupported`, `not_sent`); the same report is idempotent,
   and conflicting reports for one attempt are rejected. A late older report
   cannot change the current attempt or end its retention. `unsupported` requires
   exactly `failure: {method: "ui/download-file", code: -32601}`; other statuses
@@ -1453,34 +1564,39 @@ File continuations also use app-only `ui.act_card`:
 - `settings_cancel_download`: `operationId`, `fileId` and `attemptId`; ends
   further delivery of that snapshot, not a Host dialog or a user-saved file.
   Cancellation is refused while a local save is in progress or its outcome is
-  uncertain. After a failure recorded as `not_published`, cancellation remains
-  available within the transfer period for the current authorized request.
+  uncertain. A finished failed delivery has no further cancellation or retry input;
+  another attempt requires a new delivery card.
 - `settings_store_backup`: `operationId`, `fileId`, current `attemptId`, a new
   `saveAttemptId`, required boolean `allowPossibleCopy`, and optional
   `previousSaveAttemptId`. It is available only for that snapshot's recorded
-  unsupported result. Possible earlier delivery requires explicit true consent;
+  unsupported result and a new delivery-purpose card. Possible earlier delivery requires explicit true consent;
   otherwise false is required. Initial consent has no previous save ID. A retry
   of a confirmed no-publication failure must name the current failed save ID.
   IDs and consent are immutable per admission; duplicates do not repeat effects.
   It accepts no arbitrary path, name or bytes.
 
 These actions validate current owner, epoch, credential and exact file/attempt
-identity independently of terminal business input. Retry/read requires a live
-transfer period. A late report can record a fact after that period ends but
+identity. Effect admissions and file reads require an active purpose card and the
+original live transfer period. Closed cards grant none of these effect permissions.
+Only an exact late result report may update an already-admitted attempt's journal,
+without modifying its closed card or starting publication. A late report can record a fact after that period ends but
 cannot restore the file or authority. Host acknowledgment ends retention and
 requests cleanup; it does not prove a save location or completed disk storage.
 `isError` does not distinguish cancellation, denial or failure. A lost response
 is unconfirmed and never triggers automatic retransmission. Cleanup failure is
-reported separately and is retried through Check status or owner recovery.
+reported separately; a new operation-status card or owner recovery can settle
+known cleanup without repeating the original effect.
 Local publication closes further Host reads/retries for that snapshot. Its
-final/file-I/O reservations survive the immutable terminal snapshot result.
+file-I/O reservation lasts until the actual admitted file work ends, independently
+of the immutable card result.
 New save admissions and retries require the original live period; admitted
 file work settles even if it outlives that period. A process restart verifies
 only an existing target bound to its recorded source identity/content. It never
 publishes a missing target or restores the departed card's action authority.
 A missing or changed source ends that private transfer; a new explicit request
-is required. An existing public backup is verified on the same descriptor used
-for each range and is never truncated or removed by temporary cleanup.
+is required. Private snapshot reads verify the artifact on the same descriptor
+used for each range. Temporary cleanup never truncates or removes a saved
+local backup.
 
 Troubleshooting recovery confirms a precise SDK run and affected unsubmitted
 requests. Admission revokes their submission authority before stopping the old
@@ -1494,8 +1610,8 @@ and addresses. Historical disconnected/failed records do not prove a current
 connection or approval. The complete recorded impact remains bound to admission;
 the display does not filter it into a new wallet-eligibility rule.
 
-Endpoint changes require fully quitting/reopening every participating app to
-apply parent configuration. Restart connection service only replaces the SDK
+Endpoint changes require stopping all participating MCP connections and waiting
+for the shared parent to exit, then reopening them to apply its configuration. Restart connection service only replaces the SDK
 child. Installation environment overrides continue to take precedence. Custom
 providers affect read data quality; use trusted Sui mainnet providers.
 
@@ -1515,12 +1631,12 @@ permissions and display facts. Clients must use the returned URI/key values.
 | `suimcp://docs/mcp-tools` | API reference: tool contracts, response fields, statuses, follow-up fields, and output boundaries. |
 | `suimcp://docs/wallet-connection` | Wallet connection reference: active read context, private SDK ownership and user-approved transaction requests. |
 | `suimcp://docs/agent-behavior` | Answer playbook: user-question flows, tool selection, and response wording boundaries. |
-| `suimcp://protocols/deepbook-v3` | Protocol reference only; use MCP tool responses and `read.list_supported_protocols` for current support. |
+| `suimcp://protocols/deepbook-v3` | Protocol reference only; use MCP tool responses and `read_list_supported_protocols` for current support. |
 | `suimcp://protocols/deepbook-margin` | Protocol reference only; no margin MCP read tools or signable actions are exposed in this release. |
 
 Only allowlisted mainnet protocol references are exposed as MCP resources.
 
-Protocol resources are not runtime registries, supported-protocol lists, live liquidity sources, route recommendations, or signing-readiness signals. Use `read.get_server_status`, `read.list_supported_protocols`, concrete tool schemas, and concrete tool responses for current product support.
+Protocol resources are not runtime registries, supported-protocol lists, live liquidity sources, route recommendations, or signing-readiness signals. Use `read_get_server_status`, `read_list_supported_protocols`, concrete tool schemas, and concrete tool responses for current product support.
 
 MCP resources are runtime-facing references that connected AI clients can read.