xinbenlv

Triage and orchestrate code reviews. Analyzes PR intent, identifies touched surfaces, assesses risk, and routes to specialist skills. Does NOT perform detailed review - delegates to specialists. Supports full pipeline with "Review PR <number>" command.

Retrieve code diff from a local git commit. Use this as the first step in a code review pipeline when reviewing local commits.

Retrieve code diff from a GitHub Pull Request via GitHub API. Use this as the first step in a code review pipeline when reviewing GitHub PRs.

Submit a code review to GitHub via the GitHub API. Use this as the final step in a code review pipeline to post review findings to a PR.

Review API contracts, breaking changes, and interface consistency. Analyzes REST/RPC endpoints, event schemas, versioning, and backward compatibility. Use when reviewing public interfaces, API routes, or service contracts.

Deep codebase context analysis like Greptile. Analyzes blast radius of changes, dependency graphs, and architectural consistency. Use when reviewing changes to core utilities, shared libraries, or database models.

Review distributed systems patterns, concurrency, and resilience. Analyzes retry policies, idempotency, timeouts, circuit breakers, and race conditions. Use when reviewing async code, workers, queues, or distributed transactions.

Review configuration, secrets, and environment handling. Checks for safe defaults, secret management, feature flags, and environment parity. Use when reviewing config files, environment variables, or feature flags.

Use when turning an approved MVP List proposal into a timed agenda. Renders commute, tour, meal, rest, dish, highlight, weather, tip, and image-plan sections with Google Maps links and clear time ranges.

Use when enriching an MVP List Place Markdown record using all available public data sources. Completes schema-shaped YAML frontmatter for structured search, including restaurant top_dishes and attraction points of interest, then validates against place.schema.json.

Use when extracting candidate MVP List places from a URL, image, pasted itinerary, map link, menu link, or free-form note. Identifies restaurant and point_of_attraction places, resolves canonical references, and creates minimal schema-shaped Markdown candidates for later enrichment.

Use when selecting and ranking MVP List places from a Markdown repo to propose trip plans. Applies filters such as location, open hours, infant friendliness, visited/liked history, place type, route fit, and user preferences, then composes three candidate plans.

ModelSense — The right model for the right job. Recommends the best LLM model and effort level for any task, based on benchmark data, task analysis, and the user's configured providers. Use when the user asks "which model should I use?", "what's the best model for X?", or wants help choosing between models/effort levels.

Use when writing prose, blog posts, articles, essays, or any non-code text content — removes AI patterns and produces natural human writing

Scan all Slack channels and DMs for unresponded messages — conversations where the bot owes a reply. Use when asked to 'check unreplied messages', 'kaopu', 'kaopu scan', 'what haven't I responded to', '有没有漏回的', '靠谱检查', or any request to find dropped conversations. Produces an actionable digest with Resume/Mark Done buttons.