Minara — Personal Crypto AI CFO
Turn your agent into a hands-on crypto CFO: trade spot and perps, manage wallets, fund accounts, and run market analysis with plain-language commands.
The docs establish a fairly complete fund-safety flow: mandatory confirmation steps, ban on -y, anti-fabricated-confirmation rules, balance pre-check, contract-address verification, address-poisoning detection, analysis/trade separation, anti-loop safeguards, and a declaration that SKILL.md contains no executable code. Deductions: core function depends on the closed-source Minara CLI holding keys/credentials in ~/.minara/, data flows to api.minara.ai/MoonPay are not fully transparent, README promotes curl|bash one-click setup (supply-chain risk), publisher unverified, and permission isolation/rollback are undocumented. No red-line risk found.
Internally consistent: trigger tables, command tables, linked reference docs, an exhaustive interactive-bypass guide, plus hang detection, max-1-retry, and explicit failure reporting. Deductions: static review cannot execute key paths; several commands (limit-order create, autopilot, config) have no non-interactive mode; error handling leans on CLI output with incomplete abnormal-path coverage; no committed test suite, so the static cap of 10 applies.
Trigger semantics are precise: positive triggers (tokens, chains, trading actions), negative exclusions (no activation for pure blockchain education), stock-ticker disambiguation, and a routing gate. Deductions: environment fit is unevidenced — core function depends entirely on overseas services (minara.ai, npm, MoonPay, Hyperliquid) with no mainland-China reachability note and no Chinese-language support statement; some boundaries depend on external CLI behavior.
Well-layered docs (SKILL.md entry + per-domain references + examples), explicit version 3.0.4, a user-choice update-check mechanism, MIT license per README. Deductions: license metadata missing inside the skill frontmatter; no changelog or version history; maintenance responsibility/update path relies on git pull; no explicit known-limitations/FAQ section; referenced files like setup.md are not visible in this evidence.
Claims full trading/wallet/market coverage with a README-cited crypto-skill-bench score of 88/100 (66 passed, 0 failed), detailed command tables, and templated outputs (balance, positions, confirmation summaries). Deductions: the benchmark is self-built and self-run, not independent; static review cannot confirm outputs are directly usable; quota gates (Free-tier research) limit practical availability; static cap of 7 applies.
README cites a public benchmark repo (76 scenarios, safety gate PASS), an npm package, and a CI publish workflow, giving some traceability. Deductions: benchmark results are publisher-reported without in-repo third-party execution evidence; no committed test suite; CLI output samples are illustrative only and not statically verifiable.
- This skill executes real fund-moving trades; before use, understand how the Minara CLI custodies keys and stores credentials under ~/.minara/.
- The README's one-click install uses a curl|bash pattern; prefer manual installation and review the script first.
- Reachability of core services (minara.ai, npm, MoonPay) from mainland-China networks is undeclared and may block usage.
- The 88/100 benchmark score is self-built and self-run by the publisher, not independently verified.
- Publisher identity is unverified by the FollowSkills registry; license metadata is missing from the skill frontmatter (MIT is only stated in the README).
- Several commands (limit-order create, perps autopilot, config) have no non-interactive mode and require pty handling in agent environments.
What does this skill do, and when should you use it?
Minara Skills is the official Agent Skill from Minara, an AI-native financial OS, that lets agents like Claude Code, OpenClaw, Codex, and Hermes execute crypto trades via the minara CLI. It covers spot trading, Hyperliquid perpetual futures (with leverage and AI autopilot), limit orders, wallet deposits and withdrawals, credit-card on-ramp via MoonPay, and real-time market data with AI analysis across EVM, Solana, and Hyperliquid. The skill is a SKILL.md instruction file backed by extensive reference docs and shell scripts, and scores 88/100 on the official crypto-skill-bench benchmark. Fund-moving operations always require a two-step confirmation flow, with built-in scam-token detection and address-poisoning checks.
After reading SKILL.md, the agent maps user intent to a command table and runs minara CLI commands itself: spot swaps/buys/sells/transfers (swap, transfer), Hyperliquid perp orders and closes (perps order, perps close), limit-order management (limit-order), balance and portfolio queries (balance, assets), deposits and withdrawals (deposit, withdraw, including MoonPay card on-ramp), market data and AI analysis (discover, ask, research, including fear & greed index, BTC on-chain metrics, Polymarket), x402 protocol payments, and subscription management (premium). Every fund-moving command requires a confirmation summary shown to the user, with execution deferred to a separate reply; the skill also runs a version-check script and aborts on insufficient balance, malformed addresses, or contract addresses that don't match known canonical tokens.
- Active crypto traders: place orders from Claude Code with a single prompt like 'Buy 100 USDC worth of ETH' without touching an exchange UI
- Perps traders: manage leveraged Hyperliquid positions and multiple wallets via prompts like 'Short BTC, 10x leverage' or 'Close all positions'
- DCA and conditional-order users: set prompts like 'Buy ETH when it drops to 3000' as limit orders the agent creates and cancels
- Wallet managers: check balances and portfolio, get deposit addresses, withdraw to external addresses, or on-ramp with a credit card via MoonPay
- Research-driven users: ask for trending tokens, fear & greed index, BTC metrics, or deep research before committing to a trade
- x402 developers: pay x402-enabled HTTP APIs directly from the Minara wallet
What are this skill's strengths and limitations?
- End-to-end coverage: spot, perps, limit orders, wallet funding, card on-ramp, and market research in one skill
- Detailed safety design: mandatory two-step confirmation for all fund-moving commands, ban on -y auto-confirm, balance pre-checks, canonical contract-address verification, address-poisoning detection, and a strict analysis/trade boundary
- Quantifiable quality evidence: 88/100 on the official crypto-skill-bench (Safety 91, 76 scenarios, 66 passed, 0 failed)
- Idempotent one-click installers for Claude Code, OpenClaw, and Hermes keep onboarding cheap
- Hard dependency on the minara CLI and Minara's account system; premium subscriptions/credits exist and pricing is not stated in the source material
- Keys and funds are custodied through the Minara CLI, so you must trust this third-party infrastructure and the minara npm package
- Platform support rests on documentation claims; no independent third-party security audit is cited beyond the vendor's own benchmark
- Traditional stock tickers (AAPL, TSLA) require disambiguation — actual stocks are not tradable, only crypto context or tokenized versions
- Interactive CLI commands can hang in agent environments; the skill caps retries at one, so complex cases may still need manual intervention
How do you install this skill?
One-click for Claude Code: send your agent the prompt 'Run "curl -fsSL https://raw.githubusercontent.com/Minara-AI/skills/main/scripts/claudecode-minara-skill-setup.sh | bash" to install Minara CLI, Minara Skills and set following config. Follow the login URL when prompted.' The script installs the minara CLI, copies the skill to ~/.claude/skills/minara, and starts login. Manual: npm install -g minara@latest, then git clone https://github.com/Minara-AI/skills.git /tmp/minara-skills, then copy the skills/minara folder to ~/.claude/skills/minara. OpenClaw and Hermes have their own one-click scripts (openclaw-/hermes-minara-skill-setup.sh); OpenClaw also supports clawhub install lowesyang/minara. Authenticate via minara login --device, or set the MINARA_API_KEY environment variable to bypass login. License: MIT per the README badge and License section.
How do you use this skill?
After installation, just talk to the agent in natural language; it runs the commands itself. Example prompts: 'Login to Minara'; 'Show my deposit address'; 'Buy 100 USDC worth of ETH'; 'Short BTC, 10x leverage'; 'Buy ETH when price hits $3000'; 'What tokens are trending?'; 'Withdraw 10 SOL to [address]'. Fund-moving actions first return a confirmation summary table — you must explicitly confirm in a new message (e.g. A) Confirm) before execution; changing any parameter voids a prior confirmation. Full CLI examples live in skills/minara/references/examples.md in the repo.
How does this skill compare with similar options?
Compared with calling exchange APIs or writing your own scripts, this skill removes the engineering around signing and order management and replaces it with natural language plus built-in safety flows — at the cost of being tied to Minara's account system and CLI. The README also points to Strategy Studio (strategy.minara.ai) for AI backtesting and factor-combination strategies, a web-side extension of the same ecosystem.