Only finished work that can be re-checked from public entry points appears here. Internal plans, features that only passed local tests, and drafts still waiting for evidence are never published early.
Record real changes and the objects they affect
Every claim carries its source and when it was checked
Only confirmed records go public
Select the EveryInfra API based on input, source, and result shape by distinguishing platform data retrieval, web discovery and reading, model analysis, and challenge handling, then decide between REST and MCP, and establish acceptance criteria for each step.
Read the record →Access multi-platform data using platform, action, and params: read the capability catalog, complete minimal requests, handle asynchronous tasks, partial results, and billing, and establish traceable data records.
Read the record →Step-by-step integration of Xiaohongshu comments_batch: preparing complete note URLs, distinguishing URL count from comment count, identifying partial and missed results, deduplicating by note, and handling empty results, retries, and target billing.
Read the record →Start with live parameters for douyin.comments, read comments using minimal requests, preserve video source, comment IDs, and reply relationships, and handle limits, empty results, timestamp parsing, and repeated observations.
Read the record →Integrate TikTok profile and user_videos endpoints, separate account and video snapshots, calculate clear engagement metrics, handle missing values, and avoid treating public data as conversion results.
Read the record →Distinguish between Business Profile, Places API, and EveryInfra location review retrieval, verify location identity, language, and time filters, preserve changes to ratings and merchant replies, and clarify sample, storage, and display boundaries.
Read the record →Step-by-step integration for reading Amazon reviews, verifying country marketplaces and product identities, separating reviews, variants, and observation logs, handling deduplication and missing data, and building issue classifications with verifiable source snippets.
Read the record →Differentiate authentication, authorization, capability, parameter, and unknown results, correct requests using the current directory, and constrain retries with a decision engine that sends no network requests to prevent error handling from becoming unauthorized privilege escalation or duplicate calls.
Read the record →Starting from real-world discrepancies in platform identifiers, parameter declarations, and MCP discovery, we design catalog snapshots, contract diffs, and example checks, clarifying caching, missing fields, and release gates to avoid treating catalog alignment as full service availability.
Read the record →Separate tasks, call attempts, delivery states, and wallet events to handle pre-charge rejections, empty results, and partial credit returns without double-counting, using integer-based offline examples.
Read the record →Within systems you own and official testing mechanisms, verify types, parameters, and result shapes. Separate the acceptance of tokens, server-side validation, business operations, and costs to avoid treating return values as successful validation.
Read the record →Design a traceable reputation monitoring workflow starting with target lists, comment identities, and incremental cursors; use offline samples to demonstrate deduplication and ingestion gaps while preserving manual review.
Read the record →Choose tools for search, content reading, and cross-checking by task phase, preserving sources, timestamps, and failure states. Inspect input contracts against public catalogs rather than treating search hit counts as credibility.
Read the record →Isolate the compatibility boundaries of the SDK, base URL, and models. Validate non-streaming text requests, errors, and billing details before restoring parameters task by task, avoiding the assumption that client compatibility equals full service compatibility.
Read the record →Use a keyless discovery probe to inspect initialization, notifications, and the tool catalog, then verify the network and authentication requirements of Claude, ChatGPT, and Cursor respectively to avoid treating protocol visibility as end-user testing across all three clients.
Read the record →