Bybit AI Trading Skill
Turn any AI assistant into a Bybit trading terminal — query markets, place spot and derivatives orders, and manage positions in natural language with zero installation.
Strong credential hygiene (env vars preferred, masked display, no raw secrets in generated code, no-withdraw guidance, AI subaccount cap and 30-day expiry), update flow has path validation, SHA256 checks, and user confirmation for JS modules; mainnet writes require CONFIRM. Deductions: mainnet is the default environment (real funds by default rather than testnet), self-update pulls remote code at session start, cloud path accepts keys pasted into conversations sent to third-party servers, and OAuth lazy-fetches a remote JS executable, widening supply-chain surface.
Exhaustive failure branches: clock drift check, fail-closed signing-type selection, categorized retCode messages, read-only fallback when modules fail to load, abort-on-checksum-mismatch. Deductions: static review cannot reproduce key paths; no committed test suite or CI evidence; no systematic malformed-input handling statement, so capped at the static anchor.
Detailed trigger keyword tables, clear module routing, explicit disambiguation rules (tradfi vs trading-bot), some Chinese keywords supported. Deductions: core function depends entirely on api.bybit.com and raw.githubusercontent.com, both unreliable from mainland China; mainnet default adds regulatory risk for mainland users; no declared non-fit boundaries (restricted regions).
Versioned (1.7.1), MIT license, attributed to Bybit, well-layered modular architecture with progressive disclosure and explicit rule priority. Deductions: no changelog, no explicit maintenance ownership/update cadence, no FAQ, SKILL.md is visibly truncated mid-section, README vs SKILL.md describe the update mechanism inconsistently (VERSION file vs manifest), and the OAuth executable is not delivered in-repo.
Covers market data through derivatives, strategy orders, and copy trading; endpoint-level detail (E8 precision, UTA_* mapping, retCode handling) suggests directly executable instructions. Deductions: static review cannot verify representative outputs; usability depends on all remote modules loading; value claims (AI subaccount caps) are link-supported but unverified.
Auditable primary material (SKILL.md, manifest mechanism, help-article links) with clear fact/rule separation. Deductions: no test suite, no CI run evidence, no independent reproduction; the manifest endpoint behavior and oauth.js contents are invisible to this review, so the static ceiling applies.
- Default environment is MAINNET (real funds); users should explicitly switch to testnet on first use.
- The skill auto-fetches a remote manifest and code at session start and lazy-loads a remote JS executable for OAuth — even with SHA256 pinning this is supply-chain risk; disable auto-update in hardened environments.
- On cloud AI platforms, keys transit third-party provider servers; always use a limited-balance, Read+Trade-only subaccount.
- Core dependencies api.bybit.com and raw.githubusercontent.com may be unreachable from mainland China networks.
- This was a static review: no code was executed and trading, signing, or update behavior was not verified; no test or CI evidence accompanies the files.
What does this skill do, and when should you use it?
This is Bybit's official AI trading skill, distributed as a single MIT-licensed SKILL.md. You paste one prompt into any AI assistant, which downloads and loads the skill, then lets you drive the Bybit V5 API conversationally. It covers market data, spot and derivatives trading, Earn products, copy trading, grid/DCA bots, on-chain token swaps, and P2P fiat through on-demand modules. Security is treated seriously: mainnet-by-default with mandatory confirmations, locally computed signatures, masked key display, and an AI Subaccount flow with built-in caps.
The skill is a set of operating instructions for the AI: it fetches a versioned manifest and module files via curl from raw.githubusercontent.com, verifies SHA256 checksums, walks the user through credential setup (environment variables, .env files, or an OAuth bootstrap), then signs and calls Bybit V5 REST endpoints (e.g. /v5/order/create, /v5/position/list, /v5/account/wallet-balance) locally with HMAC-SHA256 or RSA-SHA256. Concrete operations include: querying prices, klines, orderbook depth, funding rates; placing market, limit, batch, and conditional orders; opening/closing leveraged positions with TP/SL and trailing stops; subscribing and redeeming Earn products; managing balances, transfers, and sub-accounts; creating grid/DCA/martingale bots; copy trading; on-chain DEX swaps via a quote-then-execute flow; TWAP/iceberg/chase algorithmic orders; P2P fiat trading; and a background self-update check at session start.
- An individual trader with a Bybit account who wants to tell Claude Code or Cursor 'buy 500 USDT of BTC' instead of clicking through the web UI
- A derivatives trader who needs fast funding-rate checks and 10x longs with stop-losses, requiring a structured confirmation for every mainnet write
- A semi-quant user who wants an AI to configure and create spot or futures grid / DCA bots, with parameter-range validation before creation
- A beginner who wants to practice on testnet with no real funds at risk, then explicitly switch back to mainnet
- A ChatGPT or other cloud-AI user with no local secret store, authorizing a Read+Trade-only, low-balance sub-account for trading
What are this skill's strengths and limitations?
- Published by Bybit itself, MIT-licensed, and maintained in step with the exchange's API
- Security rules are concrete, not slogans: local signing, masked key display, ban on real credentials in code blocks, capped AI subaccounts, public-key OAuth flow, fail-closed checksum verification
- Modular on-demand loading keeps context small, with graceful degradation to read-only mode on failure
- Unusually broad coverage — spot, derivatives, Earn, strategy orders, bots, copy trading, on-chain, and P2P in one skill
- Detailed error handling with domain-split retCode tables and mandatory backoff rules
- Requires you to supply your own API keys; on cloud AI platforms keys are pasted into the conversation and transit the provider's servers
- Self-update pulls and overwrites local files from a remote manifest — triple-verified, but still requires trusting Bybit's manifest endpoint
- README and SKILL.md version numbers disagree (1.5.5 vs 1.7.1), and the source shows no test suite or independent audit
- Purely prompt-driven: order correctness depends on the host model's adherence to long instructions, with real funds at stake on mistakes
- The OAuth executable is lazily fetched; skipping the Bootstrap step on fresh installs fails with Cannot find module
How do you install this skill?
No packages or CLI tools required. The official method is to send this line to any AI assistant that can read URLs: 'Please read https://raw.githubusercontent.com/bybit-exchange/skills/main/SKILL.md, save it as a skill, and help me trade on Bybit.' The AI downloads and saves SKILL.md automatically. In local tools like Claude Code you can alternatively place the repository files in your skills folder; each client loads the folder per its own Agent Skills conventions.
How do you use this skill?
1) Get an API key: preferably create an AI Subaccount in the Bybit mobile app (defaults: 5,000 USD cap limit, 30-day key expiry), Read+Trade permissions only, IP whitelisted. 2) Configure credentials: local CLIs use export BYBIT_API_KEY / BYBIT_API_SECRET / BYBIT_ENV in ~/.zshrc or ~/.bashrc; OpenClaw uses ~/.openclaw/.env; or use one-click OAuth (run the oauth module's Bootstrap first, or node commands will fail with Cannot find module). 3) Send natural-language commands like 'What's the BTC price?', 'Buy 500 USDT worth of BTC', 'Open a 10x BTC long', 'Check my balance'. Mainnet is the default and write operations require typing CONFIRM; say 'switch to testnet' to practice risk-free.
How does this skill compare with similar options?
The source names no direct competitors. Versus calling the Bybit V5 API or official SDKs yourself, this skill's edge is zero-code natural-language interaction with built-in safety rules; versus trading manually in the web app, it enables batch and conditional operations via AI. No comparative data against other AI trading skills is provided.