version: "1.0.4" name: okx-dex-strategy description: "Limit-order strategy trading on OKX Agentic Wallet. Use this skill when the user wants to place a price-triggered limit order (buy a dip, take profit, stop loss, chase a high), cancel one or more pending orders, list active or historical orders, or resume orders that have been suspended by SA TEE upgrades. Distinct from okx-dex-swap (market orders, immediate execution at the best available aggregated price). Strategy orders are stored on the Agentic Wallet TEE and execute automatically when the user-defined trigger fires. Trigger phrases: limit order, place limit order, buy at price, sell when price reaches, take profit at, stop loss at, chase high, buy dip, cancel order, cancel all orders, my orders, list orders, active orders, suspended orders, resume orders, recover suspended orders, trader mode, agentic limit order." license: MIT metadata: author: okx version: "3.3.8" homepage: "https://web3.okx.com"

Onchain OS DEX Strategy (Phase 1)

4 P0 subcommands that wrap the Agentic Wallet limit-order surface — create-limit, cancel, list, resume. SA activation (Trader Mode upgrade / re-upgrade) is performed transparently by the CLI when the BE returns UPGRADE_REQUIRED; the skill does not need to expose that detail.

Pre-flight Checks

Read ../okx-agentic-wallet/_shared/preflight.md. If that file does not exist, fall back to _shared/preflight.md. Strategy endpoints require an authenticated Agentic Wallet session — confirm login before running any subcommand.

Display labels & output language (single source of truth)

This section is the canonical rule for user-facing strings. Every other section in this skill defers to it.

Canonical Display labels — the only strings the agent may surface to the user. The CLI returns these directly (statusLabel) for status; for strategyType the agent looks them up from the §strategyType and §status tables below.

Translation rule — match the user's conversation language. Display labels above are canonical English. When the user converses in another language, the agent translates the label at output time to match the conversation language.

Never (these rules apply everywhere in this skill):

• mix two languages in one label (pick one — never render the English label and a translation side by side),
• expose the underlying enum name (BUY_DIP, CHASE_HIGH, COMPLETED, …) to the user,
• expose the underlying CLI flag value (buy_dip, chase_high, completed, cancelled, …) to the user,
• pass through the CLI's raw statusLabel verbatim when the user is conversing in a non-English language — translate it.

Notes:

• CHASE_HIGH renders as `Buy Above` in English (not "Chase High").
• SPEEDING_UP (-4) was removed 2026-05-08 — not a valid filter or display value.

Boundary vs okx-dex-swap

If the venue is named explicitly (Uniswap, PancakeSwap, Raydium, Curve, ...) → re-route to okx-dapp-discovery. This skill is for OKX-aggregated limit orders only.

Command Index

1. onchainos strategy create-limit

Place a single price-triggered limit order.

onchainos strategy create-limit \
  --chain-id <id|alias> \
  --from-token <address> \
  --to-token <address> \
  --amount <decimal-string> \
  --direction <buy|sell> \
  --trigger-price <usd> \
  [--current-price <usd>] \
  [--slippage <value>] \
  [--mev-protection <on|off|default>] \
  [--expires-in <secs>]

Output (JSON, always — the CLI has no human-format mode):

{
  "ok": true,
  "data": {
    "orderId": "<id>",
    "status": <int>,
    "statusLabel": "<label>",
    "estimatedWaitTime": <int|null>,
    "eventCursor": "<string|null>"
  }
}

Solana orders return `estimatedWaitTime=0` — the order is queryable immediately; for all other chains the agent follows §Async wait pattern (fixed 3-second sleep before re-querying).

Default order expiry

Phase 1 BE default = 7 days (604800 seconds). This is the single source of truth — everywhere else in this skill that mentions "7 days" (the §Two-step confirmation expiry note, the §list "Expire after" column note, etc.) derives from here. If BE changes the default, update this section and propagate to the other mentions.

Slippage interpretation reminder (no echo from CLI): --slippage input is in percent units — 0.05 = 0.05% (wire "0.0005"), NOT 5%. To set 5% slippage, pass --slippage 5. The skill / agent should convey this to the user at input time; the CLI itself does not echo a confirmation line.

Supported chains (Phase 1, locked)

Strategy orders are only supported on these 6 chains. Any other chain MUST be rejected upfront by the agent — do not call create-limit and do not even open the Step 1 confirmation.

Pre-flight rule (agent): when the user mentions a chain, resolve it to its chainIndex and check this list. If the chain is not in the table (e.g. Polygon 137, Optimism 10, Avalanche 43114, Linea 59144, Sui 784, Tron 195, ...), respond directly with:

Strategy orders are only supported on Ethereum / BSC / X Layer / Solana / Base / Arbitrum right now. <requested chain> is not supported — pick one of these to continue.

Do NOT proceed to Step 1 confirmation. Do NOT call the CLI. The CLI also defends against this (validates against the same 6-chain whitelist before BE), but the agent catching it earlier saves a round trip and gives a clearer message tied to the user's exact phrasing.

strategyType enum (BaseOrderStrategyTypeEnum, locked)

The Display label column is the only user-facing string — see §Display labels & output language for the cross-cutting rule. The agent never sees, computes, or passes the strategyType integer; the CLI maps everything internally.

Strategy type derivation (CLI-owned; agent uses table only for display)

strategyType is fully derived inside the CLI from (--direction, --trigger-price, current market price). The agent must NOT pass a strategy type — there is no --type flag. The same table is reproduced here so the agent can compute the Display label for the Step 1 confirmation page without re-deriving the wire integer.

Equality is folded into the "aggressive" side (CHASE_HIGH / STOP_LOSS) — same rule the CLI enforces.

Agent flow:

1. Parse direction (buy / sell) from user intent ("buy" / "ape in" / "snap up" → buy; "sell" / "take profit" / "stop loss" / "exit" → sell). Passed verbatim as --direction <buy|sell>.
2. Fetch current price — call onchainos market price --chain <chain> --address <token>, read data[0].price. For BUY direction query the to-token's current price; for SELL direction query the from-token's current price. The agent needs this for (a) Step 0 USD-value pre-flight, (b) the Step 1 confirmation page "Trigger Price vs current", and (c) computing the Display label per the table above.
3. Pass `--current-price <usd>` to the CLI so it does not re-fetch. (If the agent omits it, the CLI fetches the same value itself — correct but one extra round-trip.)

Two-step confirmation flow (Agent must follow)

create-limit is a write operation. The agent MUST present a confirmation summary to the user first and only call the CLI after the user explicitly confirms. The CLI itself does not gate (it calls BE directly); this contract is enforced at the skill layer.

Step 0 — Minimum order value pre-flight (must run before Step 1):

BE enforces a minimum order value of $1 USD (returns error 100010 ORDER_AMOUNT_TOO_SMALL otherwise). To avoid wasting a round-trip and a confirmation page on an amount that BE will reject, the agent MUST verify the from-side USD value first.

1. Fetch the from-token price (USD):

• If from-token is a well-known stablecoin (USDT / USDC / USDG / USDe / DAI / FDUSD / ...): assume from_price ≈ 1.0 without an HTTP call.
• Otherwise: call onchainos market price --chain <chain> --address <from_token>, read data[0].price as from_price.

1. Compute USD value: usd_value = from_amount × from_price.
2. If `usd_value < 1.0`:

• Compute min_from_amount = ceil(1.0 ÷ from_price) rounded up to a reasonable display precision for the token (e.g. whole units when from_price ≥ 0.1; 2-4 significant digits otherwise).
• Surface exactly this single canonical line to the user, with no extra prose — no USD-value math, no $1 threshold mention, no echo of the user's original amount, no follow-up sentence, no apology:

Minimum order amount: <min_from_amount> <from_symbol>

Translate the prefix at output time per §Display labels & output language (e.g. for a Chinese user the agent renders the same fact in Chinese). The structure stays single-line: <localised prefix> <min_from_amount> <from_symbol>.

• STOP. Do NOT render Step 1. Do NOT call the CLI. Wait for the user to provide a larger --amount, then re-run Step 0 from the top.

1. If `usd_value ≥ 1.0`: carry from_price forward (Step 1's "Value" column reuses it; no need to re-fetch) and proceed to Step 1.

Example (user wants to spend 1 OKB on a chain where OKB ≈ $0.10):

• from_price = 0.10, usd_value = 1 × 0.10 = 0.10 < 1.0 → fail
• min_from_amount = ceil(1.0 / 0.10) = 10
• Output: Minimum order amount: 10 OKB
• Stop. No Step 1, no extra prose.

Step 1 — Show the order summary for the user to confirm. Five top-level categories with sub-items; the agent may freely organise prose at runtime, but no category may be dropped:

Estimated Amount / Value formulas:

• Buy direction (BUY_DIP / CHASE_HIGH):
• Estimated Amount = from_amount ÷ trigger_price (in units of to-token)
• Value = from_amount × from_token_USD_price (if from is a stablecoin, ≈ from_amount)
• Sell direction (TAKE_PROFIT / STOP_LOSS):
• Estimated Amount = from_amount × trigger_price (in units of to-token, usually a stablecoin)
• Value = from_amount × trigger_price (if to is a stablecoin, equals Estimated Amount)

Slippage display rules:

• User did NOT mention slippage in the conversation → display Slippage: Default 15%, and omit `--slippage` on the CLI call (the CLI default is 15).
• User explicitly said "slippage X%" / "use X% slippage" / similar → display Slippage: User-specified X%, and pass --slippage X on the CLI call.

Structural example (display labels come from the §strategyType / §status tables; see §Display labels & output language for the cross-cutting rule):

1. Chain: Arbitrum
2. Order Type: Buy Dip
3. From: USDC 10
4. To:
   - Symbol: ARB
   - Trigger Price: $0.10
   - Estimated Amount: 100 ARB
   - Value: $10
5. Slippage: 15% (default)
 
If the trigger condition is not met within 7 days, this order auto-expires.
 
Reply confirm / change / cancel.

Expiry note (mandatory): After the 5 categories and before the reply prompt, the agent must surface that the limit order auto-expires 7 days after creation if the trigger never fires. Default phrasing: If the trigger condition is not met within 7 days, this order auto-expires.

Step 2 — Handle the user's reply:

• User says "confirm" / "yes" / "submit" → call onchainos strategy create-limit ... with the --type derived in Step 1.
• User says "change amount = 5" / "set trigger to 0.08" / similar → update the corresponding field and re-render Step 1 for another confirmation.
• User says "cancel" / "abort" → do NOT call the CLI; acknowledge that the order was discarded.

Hard constraints:

1. Never call `strategy create-limit` until the user has explicitly confirmed.

2. Estimated Amount / Value are agent-side estimates derived from trigger_price, not BE quotes. The realised fill amount is decided at BE execution time by slippage and aggregator routing; the agent must not present these estimates as "actual fill amounts".

3. --trigger-price is a USD price. The agent must make this clear to the user to avoid confusion with "exchange rate = X from-token per 1 to-token".

4. Never render Step 1 when Step 0's USD-value check fails. Output the single-line minimum-amount warning instead and stop — the user must restart with a larger --amount.

2. onchainos strategy cancel

Cancel a single, batch, or all active orders. Pass exactly one of the three flags:

onchainos strategy cancel --order-id <id>
onchainos strategy cancel --order-ids id1,id2,...
onchainos strategy cancel --all

Output (JSON): {ok:true,data:{updateNum:N,estimatedWaitTime:null|n}}. updateNum is the count BE accepted, not the count that reached terminal state — re-query with list after the wait.

3. onchainos strategy list

onchainos strategy list \
  [--order-id <id>] \
  [--status active,suspended,...] \
  [--chain-id 1,501] \
  [--token <address>] \
  [--limit <int>] \
  [--cursor <string>] \
  [--strategy-mode 7]

Two modes:

• Single order: pass --order-id <id> → GET openOrderDetail (returns full order shape).
• Page query: omit --order-id → POST getOpenOrder. The active wallet's addresses are auto-supplied; pass --limit (max 100, default 100) and --cursor from the previous response's nextCursor for pagination.

Flag accepts-CSV inconsistency — --status and --chain-id accept comma-separated lists (BE fields are orderStatusList / chainIdList, plural). --token accepts a single address only — BE schema 2026-05-09 replaced the prior tokenAddressList: List<String> with a singular tokenAddress: String field. For multi-token queries, the agent calls list once per token and merges the results.

getOpenOrder request body fields (locked)

status enum (TeeSaOpenOrderStatusEnum, locked)

Full 10-state mapping — agent reads data.list[].status integer and looks up the display label here:

SPEEDING_UP (-4) was removed 2026-05-08 — not surfaced and not a valid filter option. The Display label column is the only user-facing string; see §Display labels & output language. CLI's statusLabel already returns the display label, so just translate it per conversation language.

Non-terminal set (5): {-3, 0, 2, 3, 4} = CANCELLING / TRADING / CREATING / ACTIVE / SUSPENDED Terminal set (4): {-7, -2, -1, 1} = EXPIRED / CANCELLED / FAILED / COMPLETED

Default status filter (when `--status` is omitted): the CLI puts orderStatusList=[-3, 0, 2, 3, 4] into the request body — the 5 non-terminal states (CANCELLING / TRADING / CREATING / ACTIVE / SUSPENDED) — and BE applies the filter server-side, returning only matching orders. Terminal orders (Cancelled / Completed / Failed / Expired) are excluded by default. Rationale: when a user asks "show my orders" they almost always mean live ones; surfacing historical terminal records pollutes the answer.

To see terminal orders the agent must pass --status explicitly:

--status accepts comma-separated values; each entry is either an integer (e.g. 4) or a string label (active, suspended, processing, creating, cancelling, cancelled, completed, failed, expired).

Agent rendering rules (when the user asks for orders without naming a specific status)

User prompts that match this rule include: "show my strategy orders" / "list orders" / "show my limit orders" / "what orders do I have" — i.e. any general "show me my orders" intent without a status qualifier.

Steps the agent must follow:

1. Run onchainos strategy list --limit 10 (no --status) — the CLI puts orderStatusList=[-3, 0, 2, 3, 4] (non-terminal set) into the request body; BE applies the filter server-side and returns only matching orders. Always pass `--limit 10` for general "show my orders" queries; full pagination is opt-in via "next page" follow-up.
2. Render the response data.list as a Markdown table with exactly these 8 columns (locked):

Per-row mapping (order matters, no extra columns):

Sample row (Solana SOL→USDC take_profit):

```

```

Address shortening rules:

• EVM (0x prefix): first 6 chars (0x + 4) + ... + last 4 chars
• Solana / other non-prefixed base58: first 6 + ... + last 4
• Strings shorter than 10 chars: do not truncate; display verbatim

Expire timezone conversion:

• expireTime from BE is a UTC millisecond timestamp
• The agent must convert to the user's current local timezone at render time (e.g. JS Intl.DateTimeFormat, Rust chrono::Local, or equivalent)
• Fixed format: MM/DD/YYYY HH:MM:SS, 24-hour clock

1. After the table, append a single combined reminder covering pagination + status filter. Include the pagination line only when `nextCursor` is non-empty.

Canonical EN; the agent translates per §Display labels & output language:

> Showing live orders by default (10 per page). > - Reply "next page" to load more. > - To filter by a specific state, ask for orders by their Display label — e.g. Completed, Cancelled, Failed, Expired. Example: "show my completed orders" → I'll re-query with that filter.

If nextCursor is empty (no more pages), drop the "next page" bullet and keep only the status-filter bullet.

1. If the user replies "next page" / similar, re-run with --limit 10 --cursor <nextCursor> carrying the previous response's nextCursor. Render with the same table format.

1. If the user names a specific status (any of the 9 Display labels), re-run as list --limit 10 --status <label> (single value) and render the same table. Drop the status-filter bullet from the reminder; keep the pagination bullet if nextCursor is non-empty.

4. onchainos strategy resume

onchainos strategy resume                          # auto-discover all SUSPENDED + canResume=true on active wallet
onchainos strategy resume --order-ids id1,id2      # explicit

When ids are omitted, the CLI runs list filtered to status=4 and keeps only orders whose canResume flag is true; the discovered ids are then submitted to reactivate. After resume, the agent should advise the user that orders whose trigger condition was already met may execute immediately — re-query with list to confirm.

Error code → Agent action

The CLI surfaces the BE error code in human-readable form. Map each code to a recommended next step:

Match by integer code, not msg string. The msg text is for humans and may change.

Execution event codes (executionHistoryList[].code)

A separate stream from the BE error codes above. Emitted by the TEE swap-trade engine while it attempts to execute an active order. Read the latest entry first; older entries are historical context.

CLI enriches each entry directly — no sidecar table needed. For every code the CLI recognises (35+ codes covering the 3xxx active series + 2xxx legacy series), list and detail responses inject three fields next to the raw code:

For codes the CLI does not recognise (BE added a new code we haven't catalogued yet), the entry passes through unchanged — name / message / terminal are absent and the agent should surface the raw BE msg field if present, or fall back to "event code <N>".

Reading patterns

• Latest entry wins. Render the most recent executionHistoryList[] item; older entries are historical context only.
• Repeated same code. When the same code recurs every ~10s without a txHash appearing, the engine is in a soft retry loop. Surface the latest message, mention the repeat count ("this is the 5th No quote due to low liquidity event"), and ask the user whether to wait, cancel, or adjust parameters.
• `terminal=true` codes. Stop polling immediately and surface to the user; engine will not retry.
• `terminal=false` codes that repeat 3+ times. Treat as user-actionable: suggest changing slippage / amount / pair / chain, or cancelling.
• Code `0` with `txHash`. Trade succeeded; surface txHash and explorer link.

Action hints by hot code

The message field is the user-facing string. The agent additionally chooses one of these action hints based on code:

Codes outside this table: read terminal from the CLI output. terminal=true ⇒ surface and stop. terminal=false ⇒ surface and wait one more cycle.

Async wait pattern

create-limit, cancel, and resume return after the request has been accepted, not after the order reached a terminal state. The CLI surfaces estimatedWaitTime (seconds, BE-supplied — typically 0 for Solana, ~3 for BSC, ~12 for ETH-class chains).

Agent recipe (locked):

1. Run the subcommand.
2. Sleep 3 seconds. A single fixed 3-second wait is enough across all supported chains in practice — empirically validated 2026-05-11. Do not read estimatedWaitTime from the response and sleep by that value; the BE-supplied number is conservative and inflates the wait unnecessarily (especially on ETH-class chains).
3. Re-query with onchainos strategy list --order-id <orderId> to read the final state.
4. If still pending after the first re-query, surface the partial state to the user; do not loop indefinitely.

Do not run a long polling loop. The CLI is not designed for that; the BE backs the order finalisation off-chain and one targeted re-query is sufficient in normal conditions.

SA activation transparency

Trader Mode upgrade / re-upgrade is performed transparently by the CLI:

• On create-limit or resume, if the BE returns code 60018 (UPGRADE_REQUIRED), the CLI calls the Agentic Wallet attestation endpoints (getAttestDocHex → ed25519 sign with the session key → registerTeeInfo), then retries the original op once. On success the user sees Trader Mode activated. followed by the normal command output.
• The agent should not ask the user to "activate Trader Mode" first — the CLI does it on demand.
• If activation fails, the original command aborts and the activation error is surfaced. Suggest onchainos wallet status to inspect the session.

Output format conventions

All strategy subcommands always emit the standard onchainos JSON envelope {ok: true, data: { ... }} on stdout. There is no --format flag — strategy CLI is agent-facing, so structured JSON is the only output and the agent renders any user-visible tables (e.g. the §list 8-column Markdown table) from that JSON.

Phase 1 limitations