CHANGELOG.md

Changelog

All notable changes to this project are documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.4.1] - 2026-04-17

Fixed

• Crawl + extraction non-seed extension (SCR-371) — the v1.4.0 "Crawl extension priority" fix only covered the seed URL. Discovered pages still fell through to the URL-path heuristic and were saved as N.html despite a JSON body, so scrapingbee export --format csv silently dropped every non-seed page (1-row CSVs). _preferred_extension_from_scrape_params now forces "json" for --extract-rules, --ai-extract-rules, and --ai-query, so every crawled page — not just the seed — is written as N.json. The _url column in exported CSVs is also populated for every row as a side effect (the manifest now records the correct .json path per URL).

Changed

• pyproject.toml project URLs — added Changelog and Issues entries so PyPI surfaces direct links to CHANGELOG.md and the GitHub issue tracker alongside Homepage / Documentation / Repository.

[1.4.0] - 2026-04-01

Added

• tutorial command — interactive step-by-step guide to CLI features (--chapter N, --reset, --list, --output-dir).
• --confirm flag for crawl — pass --confirm yes to skip the interactive discovery-phase prompt in scripts.
• Discovery-phase warning for crawl — when --extract-rules, --ai-query, --ai-extract-rules, --return-page-text, or bare screenshot is active, crawl warns that each page will cost two requests and prompts to confirm.
• Binary URL skip in crawl discovery — crawl skips the HTML discovery re-request for URLs with known binary extensions (.jpg, .png, .pdf, .css, .js, etc.) that can never contain links.
• scrapingbee --scraping-config auto-routes to scrape command. URL is optional when using a saved config.
• scrapingbee --resume (bare) discovers incomplete batches in the current directory and shows resume commands.
• --extract-field and --fields in batch mode — extract-field works with individual files; --fields filters JSON output keys and supports dot notation across all output formats.
• --flatten-depth for export CSV to control nesting depth (default 5).
• --overwrite flag to skip file overwrite prompts.
• Batch metadata saved to .batch_meta.json for resume discovery.
• Structured user-agent headers (User-Agent-Client, User-Agent-Environment, User-Agent-OS).
• scripts/sync-skills.sh — syncs the canonical .agents/skills/ tree to all AI platform directories.
• --smart-extract — client-side extraction with path language; auto-detects JSON, HTML, XML, CSV, Markdown, and plain text.
• Path language for --smart-extract: .key, (escaped key), [0], [0:5] slicing, [0,3] multi-index, [keys]/[values], ...key recursive search, ~N context expansion, [=filter]/[key=filter] value filters, [=/pattern/] regex filters, | OR, & AND.
• JSON schema mode for --smart-extract — accepts the same format as --extract-rules for structured extraction.
• Full path language support in --extract-field and --fields.
• Auto-parse JSON strings during dot-path traversal so nested stringified JSON is traversed transparently.
• Chainable ~N context expansion — works anywhere in the path and can be chained: ...text[=*$49*]~2.h3 finds a value, goes up N levels, then continues traversal.
• [!=pattern] negation filter — exclude values or dicts matching a pattern: ...div[class!=sidebar].
• [*=pattern] glob key filter — find dicts where any key's value matches: ...*[*=faq].
• Scalar-only value matching — filters only match strings/numbers/booleans, not dicts or lists (prevents false positives).
• List flattening in recursive search — ...section[id=faq] now correctly finds individual elements, not nested lists.

Fixed

• or N falsy-value bug — --retries 0, --backoff 0, and similar zero-value numeric options now work correctly across all commands (37 occurrences).
• Crawl extension priority — --extract-rules, --ai-extract-rules, and --ai-query now correctly produce .json output files instead of .html.
• Back navigation at first step — pressing ← at the first tutorial step now shows "Already at the first step." instead of silently re-running it.
• Tutorial sort option text — CH13-S01 (Amazon) and CH14-S01 (Walmart) what_to_notice text now lists the correct valid sort values.
• Tutorial CH16-S01 now saves ChatGPT response to file with inline preview.
• Tutorial CH02-S01 and CH03-S01 now save output to file with inline preview.
• Output routing — CSV and NDJSON batch output now correctly uses --output-file. Individual files use --output-dir.
• Flag validation — conflicting flags now produce clear errors instead of being silently ignored.
• TextIOWrapper leak in CSV stdout output no longer closes stdout.
• File write errors now show clean messages instead of Python tracebacks.
• Network errors caught globally with human-friendly messages.
• Dropped batch results from async exceptions now counted as failures.
• --update-csv requires CSV input file.
• --resume requires --output-dir.
• Negative --concurrency rejected.
• --every warns when seconds are rounded to minutes.
• Auth distinguishes network errors from invalid API keys.
• Crawl project spider mode rejects API params with a helpful error showing ScrapingBeeRequest usage.
• Spider name detection fixed when --project is set (dots in spider names no longer misidentified as URLs).
• Export --columns errors with available column names when no columns match.
• "global --input-file" wording removed from all error messages.

Changed

• --output-format choices simplified to csv and ndjson. Default (no flag) writes individual files.
• Crawl concurrency defaults to 1 (with warning) when usage API fails, instead of silent 16.
• Fast-search credit cost corrected to 10 in all skill reference files (was incorrectly documented as 5).
• Skill reference docs updated: --purchased (YouTube), --save-pattern (crawl), --flatten-depth (export), content_md5 (batch output manifest).
• AI platform agent files synced across .agents/, .github/, .kiro/, .opencode/, plugins/; .github/ agent renamed from .agent.md to .md.

[1.3.1] - 2026-03-30

Added

• --scraping-config parameter for scrape and crawl commands. Apply a pre-saved scraping configuration by name from the ScrapingBee dashboard. Inline options override config settings.
• --start-page for walmart-search to paginate search results.
• --device for walmart-product to select device type (desktop, mobile, tablet).
• --purchased filter for youtube-search to filter by purchased content.
• Parameter value flexibility. Choice parameters now accept both hyphens and underscores interchangeably (e.g. --sort-by price-low and --sort-by price_low both work).
• Improved internal validation.

[1.3.0] - 2026-03-27

Added

• scrapingbee unsafe command for managing advanced feature status.
• Audit logging.

Changed

• scrapingbee logout resets all advanced feature settings.

[1.2.3] - 2026-03-25

Added

• ChatGPT --search, --add-html, --country-code flags: The chatgpt command now supports web-enhanced responses (--search true), full HTML inclusion (--add-html true), and geolocation (--country-code gb). --search false is silently ignored (only true sends the param).
• Auto-prepend https://: URLs without a scheme (e.g. example.com) now automatically get https:// prepended, matching curl/httpie behavior. Works for scrape, crawl, and --from-sitemap.
• --extract-field path suggestions: When --extract-field doesn't match any data, the CLI now prints a warning with all available dot-paths instead of silent empty output.
• Exact credit costs in --verbose: SERP commands (Google, Fast Search, Amazon, Walmart, YouTube, ChatGPT) now show exact credit costs based on request parameters (e.g. Credit Cost: 10 for Google light requests) instead of estimated ranges.
• Unit tests for all v1.2.3 changes: 39 new unit tests in tests/unit/test_v122_fixes.py plus 8 new e2e tests (FX-01 through FX-08).
• CLI documentation page: Full docs at https://www.scrapingbee.com/documentation/cli/ — installation, authentication, all commands, parameters, pipelines, and examples.

Fixed

• --allowed-domains crawl bug: Fixed a bug where --allowed-domains caused crawls to produce no output. Scrapy's built-in OffsiteMiddleware was reading the spider's allowed_domains attribute and filtering out all ScrapingBee proxy requests. Renamed to _cli_allowed_domains to avoid the conflict.
• --max-depth with non-HTML modes: Disabled Scrapy's built-in DepthMiddleware which incorrectly incremented depth on discovery re-fetches, breaking --max-depth when using --ai-query, --return-page-markdown, or other non-HTML output modes.
• Misleading screenshot warning removed: --screenshot-full-page true without --screenshot no longer prints a false "has no effect" warning — the API handles it correctly and produces a valid screenshot.
• Fast Search credit cost: Corrected from 5 to 10 credits in the estimated fallback.

Changed

• Installation recommendation: Docs now recommend uv tool install scrapingbee-cli over pip install for isolated, globally-available installation without virtual environment management.
• Version bumped to 1.2.3 across pyproject.toml, __init__.py, all skill files, and plugin manifests.

[1.2.2] - 2026-03-16

Changed

• Plugin directory restructured: Separated marketplace catalog from plugin content. Plugin now lives at plugins/scrapingbee-cli/ with its own .claude-plugin/plugin.json, matching the Claude Code marketplace spec.
• marketplace.json fixed: Moved top-level description to metadata.description, updated plugin source to ./plugins/scrapingbee-cli, removed non-spec $schema field.
• AGENTS.md upgraded: Now comprehensive and self-contained — covers all commands, options, pipelines, extraction, crawling, scheduling, credit costs, troubleshooting, and known limitations. Serves as the single source of truth for tools that read AGENTS.md (Codex CLI, Cursor, Windsurf, Amp, RooCode, Continue, and others).

Added

• GitHub Copilot skills: Added .github/skills/scrapingbee-cli/ for Copilot skill discovery.
• OpenCode skills: Added .opencode/skills/scrapingbee-cli/ for OpenCode skill discovery.
• sync-skills.sh updated: Now syncs skills to .github/skills/ and .opencode/skills/ in addition to existing destinations.

[1.2.1] - 2026-03-16

Fixed

• Marketplace plugin install: Changed "source": "." to "source": "./" in .claude-plugin/marketplace.json to match Claude Code's marketplace schema validator.

[1.2.0] - 2026-03-16

Added

• --update-csv: Fetch fresh data and update the input CSV file in-place with the latest results. Replaces the old --diff-dir workflow.
• Cron-based schedule: schedule --every INTERVAL --name NAME CMD registers a cron job. Multiple named schedules supported. Use --list to view active schedules with running time, --stop NAME or --stop all to remove them. Replacing a schedule prompts for confirmation.
• Per-command options: Options are now per-command (shown via scrapingbee [command] --help) instead of global. API-specific options are grouped (Search, Filters, Locale, etc.).
• --output-format [files|csv|ndjson]: Choose batch output format — files (default, individual files), csv (single CSV), or ndjson (streaming JSON lines to stdout).
• --deduplicate: Normalize URLs and remove duplicates before batch processing. Also available on export for removing duplicate CSV rows.
• --sample N: Process only N random items from input file — useful for testing configurations cheaply.
• --input-column: CSV input support — --input-file data.csv --input-column url reads from a named or indexed column.
• --post-process: Pipe each batch result through a shell command (e.g. --post-process 'jq .title') before saving.
• Crawl --include-pattern / --exclude-pattern: Regex filters for which links the crawler follows.
• Crawl --save-pattern: Only save pages matching this regex. Other pages are visited for link discovery but not saved. Useful for crawling through category pages to reach detail pages.
• Rich batch progress: Progress display now shows [N/total] 50 req/s | ETA 2m 30s | Failures: 3%.
• Export --flatten: Recursively flatten nested JSON dicts to dot-notation CSV columns. Lists of dicts are index-expanded (e.g. buybox.0.price).
• Export --columns: Cherry-pick CSV columns by name. Rows missing all selected columns are dropped.
• Auth validates API key: scrapingbee auth now calls the usage endpoint to verify the key before saving.
• Logout checks schedules: scrapingbee logout warns about active schedules and offers to stop them.
• Active schedule hint: Every command shows a one-line reminder when schedules are running.
• Crawl resilience: parse() catches errors from non-HTML responses (JSON, plain text) instead of crashing.

Removed

• --diff-dir: Removed from batch and export. Use --update-csv for refreshing data instead.
• --auto-diff: Removed completely.
• --daemon: Removed from schedule. Schedule now uses cron jobs that persist across sessions.
• Global flags: Options are now per-command. Removed option reorder logic and --option=value rejection.

Fixed

• Crawl broken on Scrapy 2.14: Fixed start → start_requests rename that broke the spider.
• Crawl --max-pages: Now enforced at both spider and Scrapy downloader level (CLOSESPIDER_PAGECOUNT). Counts actual fetched pages, not discovered URLs.
• Crawl JSON response crash: _extract_hrefs_from_response now catches ValueError when response is JSON instead of HTML.
• --robots.txt: Disabled ROBOTSTXT_OBEY since ScrapingBee handles robots.txt compliance.
• "Batch complete" sent to stderr: Moved back to stdout.
• CSV export for product pages: _find_main_list heuristic improved to not expand reviews/variants from single-item detail pages.
• Integration test fixes: Fixed hyphenated choice values (best-match, this-week), changed Amazon --country us to gb.

[1.1.0] - 2025-03-02

Added

• Shell-safe YouTube duration aliases: --duration short / medium / long as aliases for "<4" / "4-20" / ">20". Raw values still work (backward compatible).

• Position-independent global options: --verbose, --output-file, and all other global flags now work when placed after the subcommand (e.g. scrapingbee google --verbose "query"), in addition to before it.

• Shell-safe --extract-field dot syntax: --extract-field organic_results.url replaces the old bracket syntax (organic_results[].url). No shell quoting needed.

• AGENTS.md: Added project-root context file for tools that have no plugin/skill system. Read automatically by Amp, RooCode, Windsurf, Kilo Code, and OpenAI Codex CLI (the only mechanism Codex supports). Contains install, auth, all commands, global flags, pipeline recipes, credit costs, and troubleshooting — self-contained so no SKILL.md is needed.

• Multi-tool agent compatibility: scraping-pipeline agent is now placed in all major AI coding tool directories — .gemini/agents/ (Gemini CLI), .github/agents/ (GitHub Copilot), .augment/agents/ (Augment Code), .factory/droids/ (Factory AI), .kiro/agents/ (Kiro IDE), .opencode/agents/ (OpenCode). All use the same markdown+YAML content as .claude/agents/ (already covers Claude Code, Cursor, Amp, RooCode, Windsurf, Augment Code). Amazon Q gets .amazonq/cli-agents/scraping-pipeline.json (JSON format required by that tool).

• Multi-tool skill compatibility: SKILL.md is mirrored to .agents/skills/scrapingbee-cli/ (Amp + RooCode + OpenCode — none read AGENTS.md for skills) and .kiro/skills/scrapingbee-cli/ (Kiro — uses .kiro/steering/ for context, not AGENTS.md). Windsurf and Kilo Code are covered by AGENTS.md instead (both read it natively), so no dedicated skill directories are needed for them.

• .claude-plugin/marketplace.json: Added Claude Code plugin marketplace manifest so the scrapingbee-cli GitHub repo is recognized as a self-contained plugin marketplace. Enables users to install via Claude Code's plugin system (/plugins install scrapingbee@scrapingbee) after registering the marketplace. Declares the single scrapingbee-cli plugin with source: "." pointing to the repo root where skills/ is discovered automatically.

• --extract-field global flag: Extract values from a JSON response using a path expression and output one value per line — e.g. --extract-field organic_results.url extracts each URL from a SERP, ready to pipe into --input-file. Supports array expansion (key.subkey) and top-level scalars/lists (key). Takes precedence over --fields.

• --fields global flag: Filter JSON response output to specified comma-separated top-level keys — e.g. --fields title,price,rating. Works on single-object and list responses.

• Per-request metadata in batch manifest: write_batch_output_to_dir now writes manifest.json (alongside the numbered output files) with per-item metadata: {"input": {"file": "N.ext", "fetched_at": "<ISO-8601>", "http_status": 200}}. Enables time-series analysis for price monitoring, change detection, and audit trails.

• Enriched crawl manifest: crawl manifest.json now uses the same enriched per-item format: {"url": {"file": "N.ext", "fetched_at": "<ISO-8601>", "http_status": 200}}.

• export --diff-dir: Compare a new batch/crawl directory with a previous one and output only items whose content has changed or are new. Unchanged items (same file content by MD5) are skipped. Prints a count of skipped items to stderr.

• google --search-type ai_mode: Added the ai_mode search type to the --search-type choice list (returns an AI-generated answer).

• youtube-metadata accepts full URLs: The command now auto-extracts the video ID from full YouTube URLs (youtube.com/watch?v=..., youtu.be/..., /shorts/...), enabling direct piping from youtube-search --extract-field results.link without sed.

• Claude Skill — Pipelines section: SKILL.md now has a prominent Pipelines table at the top listing the 6 main multi-step patterns with exact one-liner commands.

• Claude Skill — Pipeline subagent: .claude/agents/scraping-pipeline.md defines an isolated subagent that orchestrates full scraping pipelines (credit check → search → batch → export) without polluting the main conversation context.

• Claude Skill — --extract-field examples added to all search command docs: fast-search, amazon-search, walmart-search, youtube-search, and google docs now include a "Pipeline" section showing how to chain into the downstream batch command.

• Claude Skill — Change monitoring pattern: patterns.md documents the --diff-dir monitoring workflow and notes that manifest.json now includes fetched_at / http_status per item for time-series analysis.

• youtube-search response normalization: The command now parses the raw YouTube API payload and outputs a clean JSON structure — results is a proper array (not a JSON-encoded string) with flat fields: link (full https://www.youtube.com/watch?v=… URL), video_id, title, channel, views, published, duration. Enables --extract-field results.link to work directly for piping into youtube-metadata.

• walmart-search → walmart-product pipeline: Search results include a top-level id field per product (e.g. "921722537"), enabling --extract-field products.id walmart-search QUERY | walmart-product — an exact parallel to the Amazon search → product pipeline. Docs updated to document this pipeline.

• Claude Skill — walmart-search → walmart-product pipeline: walmart/search.md, patterns.md, and SKILL.md pipeline table updated to document --extract-field products.id → walmart-product.

• Claude Skill — YouTube search output schema corrected: reference/youtube/search-output.md now documents the clean normalized schema (link, video_id, title, channel, views, published, duration).

• Tests: Unit tests for _normalize_youtube_search (8 tests: results array, link construction, title/channel extraction, video_id field, items without videoId skipped, already-array passthrough, invalid JSON passthrough, other fields preserved).

• Tests: Unit tests for write_batch_output_to_dir manifest writing (5 tests: correct structure, errors omitted, skipped items omitted, no manifest when all fail, screenshot subdir in manifest path).

• Tests: Unit tests for _extract_field_values (7 tests: array subkey, top-level scalar/list, missing key, invalid JSON, missing subkey items, empty array) and _filter_fields (5 tests: dict filter, nonexistent keys, empty fields, invalid JSON, list filter). Global --extract-field, --fields, and ai_mode coverage in CLI help tests.

• Tests: Unit tests for export --diff-dir (4 tests: all unchanged, changed item, new item, mixed). Unit test for new dict-valued manifest format in CSV export.

• schedule command: scrapingbee schedule --every INTERVAL CMD repeatedly runs any scrapingbee command at a fixed interval (supports 30s, 5m, 1h, 2d). --auto-diff automatically injects --diff-dir from the previous run for change detection across runs.

• --diff-dir global option: Compare batch/crawl output with a previous run — unchanged files (by MD5) are not re-written and are marked "unchanged": true in manifest.json. Works with all batch commands.

• RAG-ready chunked output: scrape --chunk-size N [--chunk-overlap M] splits text/markdown responses into overlapping NDJSON chunks (each line: {"url", "chunk_index", "total_chunks", "content", "fetched_at"}). Ready for vector DB ingestion or LLM context windows.

• Enriched batch manifest: manifest.json now includes credits_used (from Spb-Cost header), latency_ms (request timing), and content_md5 (MD5 hash of response body) per item. content_md5 powers the --diff-dir change detection.

• Estimated credit costs in verbose mode: SERP endpoints (Google, Fast Search, Amazon, Walmart, YouTube, ChatGPT) don't return the Spb-Cost header. --verbose now shows estimated credit cost from hardcoded values in credits.py when the header is absent.

• E2E test suite: 182 end-to-end tests covering all commands, batch/crawl, export, schedule, diff-dir, verbose output, and edge cases.

• Tests: Unit tests for read_input_file, crawl spider manifest fields (credits_used, latency_ms), estimated credit cost display, chunk_text, _parse_duration, schedule helpers.

• Progress counter: Batch runs now print a per-item [n/total] counter to stderr as each item completes (with (error) or (skipped) suffix when applicable). Suppress with global --no-progress flag.

• CSV export: scrapingbee export --format csv flattens JSON batch/crawl output to a tabular CSV. API responses with a top-level list (e.g. organic_results, products, results) expand to one row per item; single-object responses (e.g. product pages) produce one row per file. Nested dicts/arrays are serialised as JSON strings. _url column is added when manifest.json is present.

• Chained workflow docs: reference/usage/patterns.md now includes end-to-end pipeline recipes: SERP → scrape result pages, Amazon search → product details (with CSV export), YouTube search → video metadata, and batch SERP for many queries.

• Resume (batch): --resume global flag skips already-completed items when re-running a batch command against an existing --output-dir. Completed items are detected by scanning for N.<ext> files (.err files are not treated as complete). Applies to all batch commands: scrape, google, fast-search, amazon-product, amazon-search, walmart-product, walmart-search, youtube-metadata, youtube-search, chatgpt.

• Resume (crawl): --resume also resumes an interrupted crawl: existing manifest.json is loaded to pre-populate already-visited URLs, preventing re-fetching.

• Crawl manifest: crawl now writes manifest.json (URL → relative filename map) to the output directory when the crawl finishes, enabling resume and export.

• Sitemap ingestion: crawl --from-sitemap <url> fetches a sitemap (or sitemap index) and crawls all discovered URLs. Handles <sitemapindex> recursively (depth limit 2) and both namespaced and bare XML.

• Export command: scrapingbee export --input-dir <dir> [--format ndjson|txt] merges numbered batch/crawl output files into a single stream. NDJSON mode enriches each record with _url when a manifest.json is present; TXT mode emits # URL headers followed by page text. Output respects --output-file.

• CI: GitHub Actions workflow (.github/workflows/ci.yml) runs unit tests across Python 3.10–3.13 on every push and pull request.

• Tests: Unit tests for validate_batch_run (credit and concurrency checks).

• Tests: Unit tests for _find_main_list, _flatten_value, and export --format csv (17 tests covering flat objects, list expansion, non-JSON skipping, manifest URL enrichment, and empty-input error).

• Tests: Unit tests for _find_completed_n (nonexistent dir, numbered files, ignores .err, ignores non-numeric stems, finds files in subdirectories).

• Tests: Unit tests for run_batch_async skip-n (resume) behaviour: skipped items are marked skipped=True with empty body; empty skip_n processes all items.

• Tests: Unit tests for the crawl double-fetch discovery mechanism (parse() triggers discovery when no links; _parse_discovery_links_only() follows links without saving).

• Tests: Help-output tests for every command (youtube-search, youtube-metadata, walmart-search, walmart-product, amazon-product, amazon-search, fast-search, chatgpt, crawl, export, schedule, usage, scrape, google) — verifying key params appear in --help. YouTube choice constants tests. Global option reordering tests (15 edge cases). Total: 343 unit tests.

• Claude Skill: reference/usage/patterns.md — multi-step workflow recipes: crawl + AI extraction (Option A one-command; Option B crawl-then-batch), batch SERP pipeline.

• Claude Skill: Prerequisites section at the top of SKILL.md so AI agents install the CLI and authenticate before issuing commands.

• Claude Skill: Output schemas (truncated JSON examples) added to all API reference docs: google, fast-search, amazon-product, amazon-search, walmart-product, walmart-search, youtube-search, youtube-metadata, chatgpt.

• Claude Skill: reference/troubleshooting.md — decision tree covering empty responses, 403/429 errors, .err files, crawl stopping early, --ai-query returning null, missing output files, and proxy recommendations.

• Claude Skill: reference/batch/export.md — documents the export command and --resume flag with examples.

• Claude Skill: reference/scrape/extraction.md — documents --ai-query and --ai-extract-rules response formats with JSON examples.

• Claude Skill: reference/scrape/strategies.md — "Why use ScrapingBee instead of WebFetch or curl?" section explaining automatic proxy rotation, CAPTCHA handling, and JS rendering as reasons to prefer ScrapingBee for all web scraping tasks.

• Claude Skill: reference/crawl/overview.md — documents sitemap mode (--from-sitemap), resume (--resume), manifest.json, and the three crawl modes (Scrapy project, URL-based, sitemap-based).

Fixed

• Claude Skill: SKILL.md frontmatter version corrected from 1.3.0 to 1.1.0 to match pyproject.toml.
• Claude Skill: reference/crawl/overview.md now accurately documents the double-fetch discovery mechanism: --return-page-text (and other non-HTML options) triggers a second plain-HTML fetch for link discovery, costing 2 credits per affected page. --return-page-markdown is exempt because markdown links are extracted directly.
• Claude Skill: Removed spurious add_html / full_html reference from reference/chatgpt/overview.md (the ChatGPT command has no --add-html option).
• Claude Skill: reference/usage/patterns.md Option B uses --preset extract-links for concrete URL discovery and documents that crawl output files are numbered (no URL manifest).
• Tests: test_root_version now asserts the exact __version__ string instead of the fragile "1.0" in out substring check.

[1.0.1] - Fixed SKILL.md

Fixed

• Claude Skill: Removed invalid tags key from SKILL.md frontmatter so it validates against allowed properties (name, description, version, etc.).

[1.0.0] - Initial release

Added

• CLI for ScrapingBee API: scrapingbee with subcommands for scrape, batch, crawl, usage, auth, and specialized tools (Google, Fast Search, Amazon, Walmart, YouTube, ChatGPT).
• Space-separated option syntax (--option value); --option=value is rejected.
• Claude Skill documentation under skills/scrapingbee-cli/ for AI-assisted usage.