kira-wallet 0.1.20 → 0.2.1

package/README.md

@@ -10,130 +10,112 @@
   <a href="docs/CLI.md">CLI guide</a>
 </p>
 
-Bring the AI you already use to your wallets. Add public wallet addresses, see
-what you hold across supported EVM chains, and ask Kira to research your tokens
-or refresh your holdings. Kira runs on your computer and uses your existing
-Codex or Claude Code account.
+Kira collects and explains public wallet holdings in a local workspace. Use its
+web app to manage wallets, inspect token markets and follow research. Your
+existing AI agent can use the same records and actions through the Kira CLI or
+optional MCP connection, then open the web app beside your conversation.
 
-The 0.1.20 candidate adds optional native Codex and Claude workspaces for web research,
-files, commands and configured integrations, combined with Kira's scoped wallet
-tools. Approvals and questions appear in chat; generated files can be downloaded.
-Existing connections keep their current work mode until you select the new mode
-in connection settings. Both providers use their installed CLI and account.
-See [native workspace behavior](docs/CLI.md#native-agent-workspaces-0120-candidate).
-
-![Kira portfolio overview with network allocation and research priorities](docs/screenshots/kira-readme-overview.jpg)
+**This branch prepares 0.2.1.** Kira now supplies wallet tools and a full-window
+management app. Your agent host manages its own conversations, models, files and
+approvals. Kira does not launch a model CLI or require a model login.
 
 ## Get started
 
-**Requires Node.js 22+ and Python 3.11+. macOS is verified.** The research and
-background-process features require a POSIX environment.
-
-Install Kira:
+**Requires Node.js 22+ and Python 3.11+. macOS is verified.** Research and
+background processes require POSIX. Install Kira and open the local app:
 
 ```sh
 npm install -g kira-wallet
 kira setup
 ```
 
-Open the local app at [127.0.0.1:8787](http://127.0.0.1:8787).
+When installing a prepared release before publication, use its verified archive
+instead of assuming the npm default already contains this branch.
 
-After your first setup, open Kira with:
+1. **Connect Alchemy.** Follow the [setup guide](docs/ALCHEMY_SETUP.md). Run
+   `kira connect alchemy` in your terminal and enter the app key at the hidden
+   prompt. The key stays in a private local file. Existing environment or file
+   references can also be configured through the CLI.
+2. **Add a watching wallet.** Enter a public EVM address and an exact name.
+   Kira checks configured networks and discovered candidates, including the
+   reachable Mint Club registry. Balances appear as they are checked; prices
+   and artwork follow. Follow, pause or resume the job in Activity.
+3. **Use your agent session.** Ask a capable local agent to read `kira skill`,
+   check `kira info --json`, and use Kira's CLI. It can open the URL returned by
+   `kira start --no-open` using its own browser or app panel. A host that only
+   supports MCP can connect to `kira tools`.
 
-```sh
-kira start
-```
+Human controls remain available without an agent. `kira start` opens the app;
+`kira stop` closes its server. `kira start --read-only` disables browser actions.
+Stopping the app or ending an agent session does not cancel accepted research.
+
+## Work from any agent session
 
-It starts your local app and opens the browser. Use `kira stop` to stop it.
-For a viewer without wallet or research actions, use `kira start --read-only`.
-
-Prefer an archive? [GitHub releases](https://github.com/realproject7/kira-wallet/releases/tag/v0.1.2)
-include the same package and its checksum.
-
-1. **Connect Alchemy.** Alchemy supplies the token index across networks.
-   Open **Connect Alchemy** in Kira and follow the [simple setup guide](docs/ALCHEMY_SETUP.md).
-   Run `kira connect alchemy` in your terminal and enter your app key at the
-   hidden prompt. The key stays in a private local file. No wallet key is needed.
-2. **Add a watching wallet.** Paste a public EVM address and give it a name.
-   Kira checks every configured network and all discovered token candidates,
-   including the full reachable Mint Club registry. There is no 128-candidate
-   limit. Balances appear as they are checked; prices and artwork follow.
-   **Analyze sell routes** is optional. It estimates sales and redemptions from
-   saved holdings and does not discover more tokens. Pause and resume work in Activity.
-3. **Connect your AI.** Install and sign in to the
-   [Codex CLI](https://developers.openai.com/codex/cli) or
-   [Claude Code](https://code.claude.com/docs/en/quickstart), then choose your
-   account from **Connect a model** in Kira Chat. Select one wallet or your whole
-   portfolio, test the connection, and choose whether to allow research tools.
-4. **Ask Kira.** Start with a question, or ask it to refresh a wallet's holdings
-   or prices. Follow the job in **Activity** while it runs.
-
-Want to look around first? Try the demo:
+The CLI reads saved data even when the web server is stopped. The default
+workspace is `~/.local/share/kira-wallet`, independent of the current folder or
+agent session. Use `--data-dir` or `KIRA_DATA_DIR` to choose another workspace.
 
 ```sh
-kira --data-dir /tmp/kira-demo demo --port 8790
+kira info --json
+kira call holdings_query --arguments '{"mode":"rank","limit":5}' --json
+kira refresh 'Daily wallet' --json --idempotency-key 'daily-refresh-001'
+kira jobs wait JOB_ID --timeout 30 --json
+kira start --no-open
 ```
 
-Open [127.0.0.1:8790](http://127.0.0.1:8790). Stop it with
-`kira --data-dir /tmp/kira-demo stop`.
+`--json` makes add, holdings refresh and price refresh return an immediate job
+receipt. A receipt is not completion: read or wait for its job, then inspect the
+published result and coverage. Reuse the same idempotency key when retrying the
+same request. [Agent tools and MCP](docs/AGENT_TOOLS.md) ·
+[Packaged agent skill](skills/kira-wallet/SKILL.md).
 
-## What can I ask?
+Useful questions include:
 
-- “Which chains do I have ETH worth $10 or more on?”
-- “Find tokens I may have overlooked. Show their quantity, recorded price,
-  market venues and any available full-balance exit quote.”
-- “Review my Blast holdings across all my wallets. What is known, and where
-  is coverage missing?”
-- “Compare my saved analyses. What changed?”
-- “Refresh prices for my daily wallet, then tell me what the job actually found.”
+- “Which chains have ETH worth at least $10 in my saved records?”
+- “Show my top five recorded holdings across all wallets.”
+- “Compare these saved analyses and explain what remains unknown.”
+- “Refresh this wallet, then show the completed result in the web app.”
 
-Kira can read approved records, compare saved analyses, and start holdings or
-price research when you enable its wallet tools. It distinguishes recorded spot
-value from an execution quote. Missing prices and holdings stay unknown;
-transfer inactivity needs transfer-history evidence. Testnets and pool liquidity
-are excluded from your portfolio's USD value.
-
-![USDC markets and balances by wallet with matching Kira analysis](docs/screenshots/kira-readme-markets.jpg)
-
-*Markets, paired token logos and each wallet's balance in one view.*
+Names and symbols can match different assets. Kira identifies tokens by chain
+and contract, retains unknown prices, and separates recorded spot value from
+amount-specific sell estimates. Saved balances alone do not establish transfer
+activity. Testnets, curve reserves and pool liquidity are excluded from USD
+portfolio totals.
 
 ## Your wallets, your controls
 
-- **Watch an existing wallet** using its public address. Watching does not grant
-  Kira permission to spend.
-- **Create an OWS wallet** with **Create wallet** in the sidebar, powered by
-  [Open Wallet Standard](https://openwallet.sh/).
-  Enter the encryption passphrase in the local form. Kira connects the public
-  EVM account; keys stay in the local OWS vault.
-- **Choose what AI can see.** Wallet access can be off, limited to one wallet,
-  or cover your portfolio. Approved context is sent through your selected
-  native CLI and model account. Their data policies still apply.
-- **Keep records locally.** No Kira account or hosted portfolio database is
-  required. Chat drafts stay in memory; saving Kira chat history is opt-in.
-
-Chat currently handles research and wallet records. Signing, transfers and trade
-execution are not implemented. [Wallet onboarding details](docs/CLI.md#browser-and-local-ows-wallets).
-
-## Useful commands
-
-```sh
-kira add <public-address> --tag 'Daily wallet'
-kira refresh 'Daily wallet'
-kira prices 'Daily wallet'
-kira status
-kira stop
-```
-
-Data defaults to `~/.local/share/kira-wallet`, separate from the installed
-package. Use `--data-dir /absolute/private/directory` before a command to choose
-another workspace. For connection, model or provider help, see the
-[CLI guide](docs/CLI.md) and [viewer guide](viewer/README.md).
-
-From 0.1.4, a partial wallet analysis explains disabled discovery, missing
-credentials and network failures directly beside the holdings. After connecting
-discovery, use **Refresh holdings** to create a new analysis. **Refresh prices**
-does not discover missing tokens. Existing workspace provider settings and
-published snapshots are preserved.
+- **Manage public addresses.** Add wallets, change names, review history and
+  refresh holdings or prices in the app or through the CLI.
+- **Refresh when needed.** Provider research starts with an explicit action.
+  Opening a page or querying records does not refresh blockchain or price data.
+  The app notices saved changes made by your agent and displays job progress.
+- **Keep records locally.** Wallets, snapshots and provider references survive
+  upgrades and agent sessions. Your agent host controls which facts it sends to
+  its model service and how it stores conversations.
+- **Create an optional OWS wallet.** The local form uses
+  [Open Wallet Standard](https://openwallet.sh/) and follows its public EVM
+  account. Enter the encryption passphrase yourself; keys stay in the local
+  encrypted OWS vault. Kira has no signing or trading tool.
+
+Alchemy remains required for new holdings and price research. Token indexes
+provide candidates; on-chain reads verify balances. Unsupported indexes and
+failed reads remain coverage gaps. **Analyze sell routes** is optional and uses
+saved holdings; it does not discover additional tokens. Price refresh preserves
+recorded balances and does not recover missing holdings.
+
+DEX price references require at least $1,000 of reported pool liquidity and
+exclude extremely one-sided pools. A displayed value is an estimate, not a
+promise of sale proceeds. Observation timestamps remain attached to the facts.
+
+## Upgrading from 0.1.20
+
+Run `kira stop` before replacing the installed package, then restart the app
+after installation. This closes the viewer and leaves research jobs intact.
+Wallet names, provider references, jobs and snapshots retain their existing
+format. The embedded chat interface and its model runtimes are removed. Saved
+conversations and generated workspaces remain on disk and can be inspected with
+`kira archive`; they are not automatically sent to your new agent session.
+[Archive commands](docs/CLI.md#read-earlier-conversation-archives).
 
 ## Build and contribute
 
@@ -144,25 +126,7 @@ npm test
 python3 scripts/check-public.py
 ```
 
-From source, use `node bin/kira.cjs` in place of `kira`.
-[Local AI tool setup](docs/AGENT_TOOLS.md) ·
-[Release preparation](docs/RELEASE.md).
-
-Original code and Kira character assets are available under the [MIT license](LICENSE).
-Third-party artwork and references retain their own terms; see
-[Third-party notices](THIRD_PARTY_NOTICES.md).
-
-### Alchemy and token coverage
-
-An Alchemy connection is required for new holdings and price lookups. Alchemy
-RPC is preferred on configured supported networks; verified public endpoints
-remain a fallback where allowed. Dedicated indexes provide token candidates,
-then on-chain reads verify balances. Other configured networks still receive
-native and Mint Club registry checks, plus an available explorer index.
-Unsupported discovery and failed reads remain unknown.
-
-DEX spot prices require at least $1,000 of source-reported pool liquidity.
-Extremely one-sided pools are also excluded. This is a conservative estimate
-policy, not a claim that holdings can be sold at the displayed price. Saved
-thin-pool quotes are excluded from totals without changing their raw evidence.
-See the [Alchemy setup guide](docs/ALCHEMY_SETUP.md).
+From source, replace `kira` with `node bin/kira.cjs`.
+[Viewer guide](viewer/README.md) · [Release preparation](docs/RELEASE.md).
+Original code and Kira character assets use the [MIT license](LICENSE).
+See [third-party notices](THIRD_PARTY_NOTICES.md) for other assets and references.

package/THIRD_PARTY_NOTICES.md

@@ -4,12 +4,6 @@ Kira's original code and character assets are provided under the root MIT
 license. Third-party reference files retain their original terms. The root
 license does not relicense them.
 
-Claude native workspaces use Anthropic's official `@anthropic-ai/claude-agent-sdk`
-0.3.309 package. That dependency and its optional platform packages retain their
-upstream terms; Kira's MIT license does not relicense them. Kira explicitly uses
-the user's installed Claude Code executable and existing native account. See
-the SDK package README and [upstream SDK](https://github.com/anthropics/claude-agent-sdk-typescript).
-
 - `sources/mintclub-multitoken.sol` retains its `BUSL-1.1` SPDX identifier.
   It is a research reference, not a contract deployed by Kira.
 - `sources/uniswap-state-view.sol` retains its MIT SPDX identifier.

package/docs/AGENT_TOOLS.md

@@ -1,145 +1,117 @@
-# Recorded research tools
-
-`kira --data-dir <private-portfolio> tools` serves a read-only MCP adapter over
-newline-delimited JSON on stdin/stdout. Protocol version is pinned to
-2025-06-18. It performs no login, model turn, provider request or research
-mutation. Diagnostics stay out of stdout. The process exits on EOF.
-
-Tools: portfolio_read, wallet_read, token_read, network_read, job_list,
-job_read, snapshot_read and snapshots_compare. Arguments reject extra fields.
-Tokens use chain ID and exact contract, or `native`. Missing records remain
-unknown. Snapshots expose facts and safe source links, with fixed provider
-errors. No arbitrary files, shell commands, credentials or mutation tools are
-available. Local access to this process grants access to the selected portfolio.
-Only connect it to an explicitly chosen client and data directory.
-
-Example client-side stdio configuration, after operator approval:
-
-```toml
-[mcp_servers.kira]
-command = "kira"
-args = ["--data-dir", "<private-portfolio>", "tools"]
+# Kira tools for your agent
+
+Kira 0.2.0 provides one local application contract to the CLI, optional stdio
+MCP adapter and human web controls. It supplies wallet logic and saved facts.
+The host agent owns its model, conversation, general tools and approval flow.
+No model process is launched by Kira.
+
+## CLI first
+
+Read `kira skill` for the packaged workflow. `kira info --json` reports the
+installed version, selected data directory, capabilities, connection readiness,
+saved-data revision and active-job status. The default data directory is
+independent of the current project and conversation. Do not create a fresh
+workspace merely because a new agent session starts.
+
+```sh
+kira call portfolio_read --arguments '{"limit":20}' --json
+kira call holdings_query --arguments '{"mode":"search","query":"ETH"}' --json
+kira call wallet_read --arguments '{"wallet":"Daily wallet","offset":0,"limit":100}' --json
+kira call holdings_refresh --arguments '{"wallet":"Daily wallet","idempotency_key":"refresh-daily-001"}' --json
 ```
 
-Replace the placeholder locally. Do not commit portfolio paths or credentials.
-Kira does not edit the operator's Codex configuration automatically.
-
-## Connected Kira chat
-
-The 0.1.20 candidate offers explicitly selected native Codex and Claude work modes.
-It exposes the existing `ResearchTools` catalog through a private per-session
-MCP bridge, with the same wallet scope and durable research-job implementation.
-The native agent chooses its own reads and general tools; it does not use the
-recorded chat's forced read plan or eight-call loop. Each tool request is checked
-against the live turn and current permissions. Refresh retries within a turn
-reuse the durable request identity. No model or native child receives provider
-credentials or portfolio-directory configuration through the bridge.
-
-The bridge accepts only authenticated loopback requests without an Origin
-header and is disabled between turns. It is distinct from the public read-only
-`kira tools` adapter above. Native sessions preserve configured user integrations
-and use a separate output workspace. Wallet scope controls Kira tools; it is not
-an OS restriction on the native CLI. See [work modes](CLI.md#native-agent-workspaces-0120-candidate).
-
-The following describes recorded wallet chat, retained for existing connections
-and available when native work mode is not selected.
-
-The in-app conversation uses a supported installed Codex or Claude CLI account.
-A generic response check is required before saving a selected provider, model,
-wallet scope and research permission. It sends no wallet facts. Authentication
-stays with the installed CLI. No desktop tokens or authentication files are copied.
-
-With research enabled, a bounded host-mediated loop provides portfolio_read,
-wallet_read, token_read, holdings_query, snapshots_list, snapshot_read, snapshots_compare,
-job_list, job_read, holdings_refresh and prices_refresh. Every operation is
-scoped to registered wallets allowed by the conversation. Historical reads use
-normalized financial projections. Unknown values and historical quote times
-remain explicit. Native shell, file, browser and MCP execution stay disabled.
-
-Automatic context and portfolio_read include every approved wallet with a
-bounded first page of up to 12 holdings. asset_count and assets_omitted retain
-the recorded inventory size. wallet_read returns a compact page of up to 100
-holdings; offset and limit are optional. Follow next_offset until null before
-claiming a complete inventory. snapshot_read and snapshots_compare also accept
-offset and limit. token_read supplies detailed curve, pool and exit evidence.
-Context-limit errors have failed tool receipts, rather than completed receipts.
-
-Chain evidence includes explicit native balances, including observed zero,
-their observation times and blocks, and registry scan blocks. Empty positive
-holding inventories do not remove these observations. Saved-analysis reads
-also retain snapshot price-reference values and times without provider URLs
-or debug fields. A comparison flag can reflect observation or reference changes
-without proving a balance change or a different set of checked networks.
-Research job reads include the approved wallet identity directly.
-
-Refresh tools admit typed durable engine jobs. Current scope and cancellation
-are checked immediately before admission. Queued work is visible in Activity;
-chat cancellation does not cancel engine work already accepted. Duplicate
-refresh requests within one turn reuse the same job. At most eight tool calls,
-420 seconds and bounded context are permitted per turn. Existing configurations
-keep their research permission off until explicitly enabled.
-
-OWS create/list and browser wallet connection are human interfaces, not model
-tools. The agent never receives passphrases, mnemonics, private keys or provider
-credentials. Messages and approved wallet facts reach the selected model service
-through the user's CLI; local storage does not imply an offline model.
-
-See [the conversation contract](ONBOARDING_CHAT_CONTRACT.md).
-The standalone stdio MCP adapter above remains read-only and separate.
-
-Native responses select the final Codex agent message. Older verified event
-streams without a phase use their last agent message. Commentary cannot become
-a tool request by concatenation. A mixed or malformed request gets one bounded
-correction; no tool executes from it and repeated malformed protocol fails
-without adding raw JSON to conversation history. Turn receipts expose only
-bounded tool names and completed/failed outcomes.
-
-Job reads include analysis phase, parent/continuation IDs and real progress
-events. Normalized chain discovery completeness is separate from a job's
-on-chain checkpoint status. Deferred candidates and pending networks are not
-failed reads or zero holdings. Historical observations cannot establish current
-execution, transfers or realized profit.
-
-Before answering, scoped research chat must declare a `records_read` plan
-containing every requested source, up to 32 declared read tools. An empty plan
-is for greetings or action-only requests. Planning cannot start a refresh.
-Each child has its own host receipt. The declared plan and literal JSON read
-requests are matched against actual arguments, including chain and contract.
-An incomplete delivered plan receives one bounded continuation attempt before
-the turn is marked partial. Full inventory
-requests require contiguous delivered pages through `next_offset: null`.
-Failed or undelivered pages remain unread, with retry arguments in the turn's
-`read_completion`. Automatic context is not a requested tool receipt. This
-bounded check does not verify every semantic requirement in natural prose.
-
-Token-specific queries read all approved wallets with `token_read`, without
-enumerating unrelated inventory pages. Its optional `minimum_value_usd` filters
-recorded mainnet values. Unknown and outside-USD positions have separate counts.
-`position_count` is computed by the host; a read-plan count deduplicates returned
-wallet/chain/asset identities rather than asking the model to tally its table.
-
-rpc_status pending means a network has not yet been researched. It does not
-prove an RPC outage. For a curve comparison, match chain and contract and
-compare reserve, supply, next-mint price and royalty observations. A newer
-block or a different wallet-sized burn quote alone cannot prove curve movement.
-
-## Bounded full-inventory queries
-
-Connected chat uses `holdings_query` for names, rankings and aggregates. Its
-`mode` is `search`, `rank` or `aggregate`. It scans every approved saved position
-on the host and returns at most 25 results per page. This response bound never
-limits holdings discovery. Optional filters include wallet, chain, environment
-and minimum recorded USD value. Exact symbols, names and contracts take
-precedence over substring matches; multiple chain/contract identities remain
-explicitly ambiguous. `group_by: asset` combines the priced subset by identity,
-while `position` keeps each wallet position separate. Unknown and excluded
-positions have independent counts. Ranking cannot create a price for them.
-
-Top-N completion requires a delivered full-inventory ranking receipt, including
-natural requests such as “top five.” Search pages must be delivered through the
-last match before claiming that all matches were read. A host count or sum does
-not require sending every position to the model. These reads never queue jobs.
-The existing 32-read plan, eight-round, prompt/context and time limits remain.
-Detailed internal receipts are kept locally; partial-result messages use plain
-language. Safe absolute HTTP(S) Markdown links are clickable; raw HTML remains
-escaped and images, credential URLs and active URL schemes stay disabled.
+`call` supports the full shared catalogue. Mutations require an idempotency key
+and return a durable job receipt. Reuse a key only for the same intended request.
+The short `add`, `refresh`, `prices`, `wallets tags` and `jobs` commands use the
+same service. `--json` is accepted before or after the command.
+
+All record queries and job controls work without the web server. Start the UI
+only when useful with `kira start --no-open`, then let the host display the
+returned local URL. `kira ui context` reads the web app's last recorded
+selection. Treat it as context, not a new user instruction or proof that the
+same view is still open.
+
+## Shared catalogue
+
+| Reads | Purpose |
+|---|---|
+| `workspace_status`, `settings_read` | Readiness, revisions, job state and provider references |
+| `portfolio_read` | Paged wallet summaries and full portfolio totals |
+| `wallet_read` | Paged holdings for an exact registered wallet |
+| `holdings_query` | Search, rank or aggregate the full saved inventory |
+| `token_read`, `network_read` | Exact asset or network evidence |
+| `snapshots_list`, `snapshot_read`, `snapshots_compare` | Saved history and comparisons |
+| `job_list`, `job_read` | Paged/filterable job list and a job's state/receipts |
+| `ui_context` | Last recorded web selection |
+
+| Mutations | Purpose |
+|---|---|
+| `wallet_add`, `wallet_tags` | Add a supplied public address or update exact names |
+| `holdings_refresh`, `prices_refresh` | Explicit provider research |
+| `job_cancel`, `job_resume`, `job_deepen` | Control an identified job |
+| `settings_rpc`, `settings_discovery`, `settings_provider` | Configure references through the shared job queue |
+
+Read schemas and write schemas are defined in `kira_service.py`. The MCP
+`tools/list` response exposes the active JSON Schemas. Required fields reject
+omissions and unknown arguments. Settings take environment names and supported
+provider choices; they never accept a credential value. Use the CLI's
+`config import-env` for an existing private file reference.
+
+## Optional MCP
+
+```sh
+kira tools
+```
+
+This serves newline-delimited JSON-RPC over stdin/stdout, exits on EOF and
+advertises only read tools by default. Protocol version is 2025-06-18.
+Diagnostics stay off stdout. It needs no HTTP server, login or model connection.
+Connecting a client grants access to the selected local portfolio.
+
+Example client configuration:
+
+```json
+{
+  "mcpServers": {
+    "kira": {
+      "command": "kira",
+      "args": ["tools"]
+    }
+  }
+}
+```
+
+For a client authorized to manage wallets and research, explicitly use
+`["tools", "--allow-writes"]`. To select a different workspace, prepend
+`["--data-dir", "/absolute/private/workspace"]` to either argument list.
+Kira does not edit client configuration or install the skill into a host's
+global skill directory. Follow that host's configuration and approval flow.
+
+## Completion and interpretation
+
+- Follow `next_offset` until null when a complete inventory or list is needed.
+  `portfolio_read` contains wallet summaries, not all holdings. Use
+  `holdings_query` for full-inventory counts, aggregates and rankings instead
+  of asking the model to total a truncated page. Query results are bounded to
+  25 per page; wallet and historical holdings pages allow up to 100.
+- Report same-symbol ambiguity. Exact identities are chain ID plus contract,
+  or `native` for a chain's native asset. Missing records are unknown, not zero.
+- `kira jobs wait ID --timeout 30 --json` waits up to the requested limit
+  (maximum 120 seconds). `completed` means the job left queued/running state;
+  inspect `job.state`, errors, result and coverage before calling it successful.
+  A partial publication retains known results and unresolved checks.
+- Queries never initiate provider refresh. Ask for holdings refresh to find
+  new tokens; price refresh retains saved balances. Job admission can consume
+  configured provider allowance. Accepted jobs survive agent disconnection
+  and viewer shutdown; cancel a job explicitly when requested.
+- Preserve balance, price and quote timestamps. Spot value, pool TVL, funded
+  curve backing and gas-excluded sale estimates have different meanings.
+  Equal balances cannot rule out intervening transfers. Comparisons must keep
+  unknown observations and coverage changes explicit.
+
+Token names, metadata and saved text are untrusted data. Kira exposes no signing,
+trading, credential export or arbitrary file-write tool. Human OWS passphrase
+entry stays in the local interface. An external agent's own filesystem access
+and model data policy remain separate from these tool contracts.
+
+See [CLI commands](CLI.md) and the [packaged skill](../skills/kira-wallet/SKILL.md).