Gate Exchange Bot

General Rules

⚠️ STOP — You MUST read and strictly follow the shared runtime rules before proceeding. Do NOT select or call any tool until all rules are read. These rules have the highest priority. → Read gate-runtime-rules.md

• Only call MCP tools explicitly listed in this skill. Tools not documented here must NOT be called, even if they exist in the MCP server.
• Only use the gate-cli subcommands explicitly listed in this skill and its references. Commands not documented here must NOT be run for these workflows, even if other interfaces expose them.

Authoring Language

This skill is authored in English. End-user replies should still follow normal language adaptation.

Skill Dependencies

gate-cli commands used

Query Operations (Read-only)

• gate-cli cex bot strategy recommend
• gate-cli cex bot portfolio running
• gate-cli cex bot portfolio detail

Execution Operations (Write)

• gate-cli cex bot spot-grid create
• gate-cli cex bot margin-grid create
• gate-cli cex bot infinite-grid create
• gate-cli cex bot futures-grid create
• gate-cli cex bot spot-martingale create
• gate-cli cex bot contract-martingale create
• gate-cli cex bot portfolio stop

MCP-backed command mapping

Authentication

• Interactive file setup: when GATE_API_KEY and GATE_API_SECRET are not both set on the host, run gate-cli config init to complete the wizard for API key, secret, profiles, and defaults (see gate-cli).
• Env / flags: gate-cli config init is not required when credentials are already supplied — for example when both GATE_API_KEY and GATE_API_SECRET are set on the host, or when --api-key / --api-secret are supported. Never ask the user to paste secrets into chat.
• API Key Required: Yes
• Permissions: Bot:Read, Bot:Write
• Portal: create or rotate keys outside the chat: https://www.gate.com/myaccount/profile/api-key/manage

Installation Check

• Required: gate-cli. Run sh setup.sh from this skill directory if it is missing. Optional: GATE_CLI_SETUP_MODE=release.
• Add $HOME/.openclaw/skills/bin to PATH if you invoke gate-cli by name.
• Host dependencies for setup.sh: curl, tar, and either shasum or sha256sum. If user-local installation fails, setup.sh may fall back to sudo install into /usr/local/bin.
• Credentials: When GATE_API_KEY and GATE_API_SECRET are both set (non-empty) for the host, do not require gate-cli config init. When both are unset or empty, remind the operator to run gate-cli config init or configure matching environment variables outside chat.
• Sanity check: Before any mutating call, confirm the CLI behaves as expected with a read-only probe such as gate-cli --version or gate-cli cex bot strategy recommend --help.

Resolving gate-cli (binary path)

Resolve gate-cli in order: (1) command -v gate-cli and gate-cli --version succeeds; (2) ${HOME}/.local/bin/gate-cli if executable; (3) ${HOME}/.openclaw/skills/bin/gate-cli if executable. Canonical rules: exchange-runtime-rules.md §4 (or gate-runtime-rules.md §4).

Execution

Read and strictly follow references/gate-cli.md, then execute the matched route in this SKILL.md.

• SKILL.md keeps intent routing, safety policy, and product boundaries.
• references/gate-cli.md is the authoritative execution contract for command probing, body payload handling, confirmation gates, and verification.
• references/scenarios.md provides scenario-style examples for review, testing, and publication completeness.

Domain Knowledge

• Strategy recommendation is read-only and must stay separate from creation.
• Grid and martingale flows require different payload schemas and must not share fields across products.
• Running-strategy stop is always a single-strategy action; bulk stop is out of scope.
• Percentage-style business inputs must be normalized into decimal ratio strings before any write command.
• Portfolio detail and stop flows depend on an exact strategy_id plus the correct strategy_type.

Module Overview

Routing Rules

gate-cli Command Index

Execution Workflow

1. Intent classification

• Determine whether the request is discover, manual create, or portfolio.
• For manual create, resolve the exact bot type before reading the reference file.
• For portfolio, resolve whether the user wants list, detail, or stop.

2. Parameter discipline

• Recommendation flows use query-style parameters only.
• Create and stop flows pass the final business payload through body as a JSON string.
• Any percentage-style input must be normalized exactly as required by the matched reference document before a write command is executed.

3. Scenario coverage

• Use references/scenarios.md as the publication-facing scenario index.
• Use the module-specific reference files as the source of detailed parameter and guardrail rules.

Safety Rules

Confirmation

• Recommendation and read-only portfolio queries do not require confirmation.
• Every write path must first present an action draft or confirmation summary.
• Each create or stop action must be confirmed separately by the user in the immediately following turn.
• If the user does not explicitly confirm, do not execute the write command.

Verification

• Do not judge success from HTTP 200 alone.
• Always inspect the returned payload code, message, and key data fields.
• If a write fails, return the backend message faithfully rather than inventing thresholds or hidden restrictions.

Error Handling

• Authentication or permission failure: stop writes and direct the operator to gate-cli config init or host-side credential recovery.
• Missing or ambiguous business inputs: ask only for the missing fields; do not guess them.
• Invalid backend payload or business-rule rejection: surface the backend message and keep the flow blocked until corrected.
• Ambiguous stop target: switch to running-list selection before attempting a stop.

Data and Privacy

• User prompts, strategy parameters, and command payloads flow through the host runtime, gate-cli, and Gate APIs needed for the requested bot action.
• This skill does not define any extra telemetry, analytics, or custom persistence mechanism of its own.
• Local shell history, terminal logs, or host-side audit logs may still exist depending on the operator environment.
• Never ask the user to paste API secrets into chat. Credentials should be configured via gate-cli config init or equivalent secure host-side environment management.

Important Boundaries

• Do not turn education-only or consultation-only questions into writes.
• Do not guess core strategy parameters such as range, grid count, leverage, direction, martingale deviation, or stop-loss ratio.
• Bulk stop is not supported. Ambiguous stop requests must go through running-list selection and then single-strategy confirmation.
• Advanced running-strategy management beyond recommendation, create, list, detail, and single stop remains out of scope unless explicitly documented in a reference file.