FlyAI Travel Search Skill
Lets Claude Code, OpenClaw, and other skill-compatible agents search flights, trains, hotels, and attractions in natural language — with bookable links, no browser tabs needed.
The README describes read-only search commands returning JSON with booking links, a data-flow diagram (agent → CLI → Fliggy MCP API), so external effects and data-flow transparency earn 2. But the repo contains only README and LICENSE — no SKILL.md, no CLI source: least-privilege policy, user confirmation, how FLYAI_API_KEY is stored/transmitted/scrubbed, dependency inventory and security posture, uninstall/rollback, and the true Alibaba/Fliggy affiliation are all unevidenced, scored 1 or 0 with those deductions.
README is internally consistent (eight commands; param tables match examples; self_consistency=2). However the actual CLI (@fly-ai/flyai-cli) is not in the repo; Node.js requirements, degraded behavior without an API key, and error handling are described in one line only, so dependency_availability and failure_messages drop to 1.
Audience (terminal/agent users), scenarios (flights, trains, hotels, POIs, Marriott), bilingual triggers are well covered (audience_and_scenarios=2; environment_fit=2 with Node.js runtime and install paths). But 'intent activation at priority 90' is a bare assertion with no trigger-word table or matching rules, and capability boundaries (booking/payment support, geographic coverage) are unstated — deducted.
README is well structured (quick start, command tables, examples, use cases, MIT LICENSE file present; install_notes=2, examples=2, license=2). Missing: CHANGELOG/version history (badge says 1.0.14 with no record), known-limitations section, a maintenance/support path beyond a DingTalk QR code; the 'alibaba-flyai' name sits on an unverified publisher and installation is fragmented across npm, GitHub and clawhub, so naming_stability, versioning_changelog, known_limitations and maintenance_responsibility are deducted.
Output contract is explicit (single-line JSON to stdout, errors to stderr, pipeable; output_usability=2). Aggregating real-time Fliggy inventory inside an agent is genuine differentiation (marginal_value=2); zero-config start plus optional key is a reasonable cost/benefit story (cost_benefit=2). Deduction: 'enhanced results' and booking links are unverified claims; actual value depends on an invisible remote service.
Commands and flags are checkable against the README itself, but key claims (official Fliggy integration, AI semantic search, real-time inventory, npm download counts) have no in-repo code, tests, or third-party evidence (claim_traceability=1); no second source corroborates anything (cross_source_corroboration=1); marketing language is mixed with factual statements without separation (fact_inference_separation=1). Static review cannot execute anything; confidence is low.
- The repo contains no source code or SKILL.md; CLI behavior, permission scope, and data flow cannot be statically verified — audit the npm package @fly-ai/flyai-cli before installing.
- Storage and transmission of FLYAI_API_KEY are undocumented; confirm it is not logged or uploaded.
- The 'alibaba-flyai' publisher is unverified; claimed Alibaba/Fliggy affiliation is assertion only and should not ground trust.
- No CHANGELOG or known-limitations section; claims like 'priority 90 auto-activation' and 'enhanced results' are unverified.
- All results come from a remote Fliggy API; verify prices and availability independently before booking.
What does this agent do, and when should you use it?
FlyAI (alibaba-flyai/flyai-skill) is a travel search skill that connects Fliggy — Alibaba Group's travel platform — to Claude Code, OpenClaw, and other skill-compatible agents. It ships with an npm-installed CLI (@fly-ai/flyai-cli) that runs on Node.js and emits single-line JSON to stdout, with errors and hints on stderr. The skill exposes eight commands: keyword-search, ai-search, search-flight, search-train, search-hotel, search-poi, search-marriott-hotel, and search-marriott-package, covering natural-language keyword discovery, semantic search, and structured flight/train/hotel/POI/Marriott queries. Every result includes direct booking links, and the skill works with zero configuration, with an optional FLYAI_API_KEY for enhanced results. End to end, an agent query activates the skill, flyai-cli calls the Fliggy MCP API, JSON flows back, and the agent renders it as Markdown. It is released under the MIT license.
Once installed, the skill performs intent-based activation at priority 90, automatically routing travel queries to /flyai commands or the flyai CLI. Its eight commands: keyword-search runs a cross-category keyword search over flights, trains, hotels, tickets, cruises, visas, and SIM cards; ai-search performs semantic search for complex intents (e.g., '3-day Hangzhou trip for Labor Day, budget 2000 per person, stay near West Lake'); search-flight does structured flight filtering via --origin/--destination/--dep-date and more (direct/connecting, cabin class, price cap, departure-hour windows, 8 sort modes); search-train offers the same for train tickets with seat classes like second class and hard sleeper; search-hotel filters hotels by destination, star rating, bed type, price, and nearby POIs; search-poi finds attractions by city across 30+ categories (nature, historical sites, museums, etc.); search-marriott-hotel and search-marriott-package query Marriott Group hotels and packages such as afternoon tea and spa deals. All queries go through the Fliggy MCP API and return JSON with booking links.
- A developer deep in a terminal session with an AI coding agent who wants a weekend-getaway plan from Shanghai — flights, beach hotel, and activities — in one keyword-search
- A budget traveler using search-flight's --max-price and --sort-type 3 to find the cheapest direct Beijing-to-Bangkok flight across a May date range
- A group-trip organizer filtering Sanya hotels with search-hotel by 4 stars, twin beds, under ¥500/night, sorted by rating
- A rail traveler using search-train with --journey-type 1 and 'second class' to find a direct Beijing–Shanghai high-speed train
- A Marriott loyalist using search-marriott-package to find afternoon tea or spa packages in Shanghai sorted by price
- A trip planner using ai-search to state budget, dates, and location preferences in one sentence and get mixed flight-and-hotel results
What are this agent's strengths and limitations?
- Searches real Fliggy inventory (Alibaba's travel platform) and returns direct booking links, so you can go from search to booking
- Eight commands span the full travel lifecycle — flights, trains, hotels, attractions, Marriott hotels and packages — across keyword, semantic, and structured query modes
- Zero-config start with no API key required; standardized single-line JSON output makes results easy to pipe or render
- Intent activation at priority 90 means travel queries in Claude Code / OpenClaw route to FlyAI without manual command invocation
- Core data depends on a single provider — the Fliggy MCP API — so service changes or outages directly affect functionality
- Deeper use requires a Node.js runtime and a global npm CLI; Claude Code additionally needs the skill directory manually copied into ~/.claude/skills
- Enhanced results rely on an optional FLYAI_API_KEY whose acquisition and billing are not documented in the source
- Many hotel/POI/seat filter values use Chinese enumerations (e.g., 大床房, 历史古迹), which favors Chinese-language use; English users must learn these conventions
- Beyond the README there is little auditable evidence (no tests or release notes shown), so real-world stability needs your own validation
How do you install or deploy this agent?
1) Install the skill — OpenClaw: clawhub install flyai or npx skills add alibaba-flyai/flyai-skill; Claude Code: cp -r /path/to/flyai-skill/skills/flyai ~/.claude/skills/flyai. 2) Install the CLI: npm i -g @fly-ai/flyai-cli (requires Node.js). 3) Verify: flyai keyword-search --query "things to do in Tokyo" should print structured JSON. 4) (Optional) configure: flyai config set FLYAI_API_KEY "your-key" for enhanced results; the skill works out of the box without any API key.
How do you use this agent?
In Claude Code or OpenClaw, simply ask naturally (e.g., 'direct flights Beijing to Shanghai next Friday under ¥600') — the skill auto-activates on travel intent; or invoke explicitly: /flyai search-flight --origin "Beijing" --destination "Shanghai" --dep-date 2026-04-25 --max-price 600. You can also use the CLI directly, e.g., flyai search-hotel --dest-name "Hangzhou" --poi-name "West Lake" --check-in-date 2026-04-10 --check-out-date 2026-04-12. Full parameter tables for each command are in the repository README; results are single-line JSON on stdout that you can pipe, filter, or have the agent render as Markdown.
How does this agent compare with similar options?
The README names no direct competitors, but the skill is positioned as the official way to bring Fliggy into skill-compatible agents like Claude Code and OpenClaw, sparing you from wrapping travel APIs and writing your own intent routing.
FAQ
Do I need to pay or provide an API key?
flyai config set FLYAI_API_KEY "your-key") unlocks enhanced results, but the key's acquisition and billing are not explained in the README.