Skip to content

The agent that turns websites into APIs!

38.2 KB
Raw Blame

Changelog

All notable changes to this project will be documented in this file.

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

[Unreleased]

[0.14.0] - 2026-09-23

Security

  • Refresh vulnerable website transitive dependencies, including js-yaml, nanoid, undici, brace-expansion and Miniflare's Sharp dependency, with patched lockfile versions and overrides.
  • Upgrade the website to Next.js 16.3.3 and its Sharp image-processing dependency to 0.35.4 to address known vulnerabilities.

Added

  • PowerShell 7+ output: generate an api_client.psm1 module and Example.ps1, with CLI execution through pwsh on Windows, macOS, and Linux.
  • Live client validation: an opt-in local test captures a fixture API, generates a Python or PowerShell client with a verified free OpenCode model, and checks fresh API data on replay. The manual live_windows CI option validates both languages using preinstalled Chrome and cached Python/npm dependencies, without downloading Chromium. Regular CI requires real PowerShell execution and CLI startup checks across all three operating systems.

Fixed

  • Windows encoding and launch compatibility (#134): read prompts and persisted text as UTF-8; accept optional UTF-8 BOMs in configuration, history and HAR files; retain readable message records around corrupt lines and replace session history atomically to preserve the previous file if saving is interrupted. Preserve Unicode client output, including Python dependency-install retries, and report unsupported batch launch configurations accurately in JSON output.
  • Resolve Chrome profiles for Windows and Linux, pass npm/npx arguments through native Node entrypoints, and use the Maven project working directory to avoid passing special directory characters through batch arguments. Unsafe nonstandard batch launchers fail explicitly.
  • Stop the managed OpenCode process tree on Windows so shutdown releases its listening port. Add native Windows regression coverage and dedicated Chrome capture/shutdown plus Python/PowerShell generation and fresh-response replay checks.
  • Expose unexpected agent capture tracebacks through RAE_DEBUG=1 on stderr, without adding diagnostics to JSON/NDJSON stdout.
  • Keep the website header pinned to the viewport during rapid scrolling on the homepage and documentation pages, while preserving the content offset. Add browser regression coverage to the website deployment workflow.
  • PowerShell execution validation: reject unsupported module names, missing Example.ps1, and script arguments that the example cannot forward, rather than silently running a different command. Generation instructions require portable module imports and terminating errors so failed API calls produce a nonzero exit status.
  • Refresh the locked Claude Agent SDK and MCP dependencies so the development environment also uses MCP 2.x and httpx2, without the legacy httpx, httpx-sse, and httpcore packages. Add subprocess smoke tests for CLI startup with legacy HTTP client imports blocked.
  • Restore the strict typing fixes from #125 and make the mypy CI job blocking. The original PR landed on an intermediate branch after its promotion PR had already merged, so the changes had not reached main.
  • Raise security minimums to Click >=8.3.3, aiohttp >=3.14.3, cryptography >=50.0.0, AnyIO >=4.14.2, PyJWT >=2.13.0, python-multipart >=0.0.31, Starlette >=1.3.1, HTTPX2 >=2.12.0 (which requires the patched HTTPCore2), and Markdownify >=0.14.1 / SoupSieve >=2.9.0 (collector extra). Export these floors in package metadata so upgrades cannot retain vulnerable versions that only the repository lockfile excluded.
  • Report agent provider runs that produce no result as failures: JSON/NDJSON emits an error result, and plain output prints the error to stderr; all exit with code 1. Preserve Claude/Cursor Ctrl+C as interrupted in non-interactive mode. Interactive Ctrl+C keeps exit code 0 and any completed partial result, including when the signal cancels an asyncio task. Programmatic cancellations propagate to asyncio callers and produce a CLI failure with exit code 1, while preserving partial results.

Compatibility

  • The secured cryptography dependency no longer supports Intel macOS or 32-bit Windows. Use Linux (including on Intel hardware), Apple Silicon macOS, or 64-bit Windows. Installing on the retired platforms may attempt an unsupported source build requiring Rust and OpenSSL development tools; this is not a supported installation path. Do not downgrade cryptography to restore old wheels. See the upstream platform changes.
  • Click now requires version 8.3.3 or newer to fix PYSEC-2026-2132. LiteLLM releases that pin click==8.1.8 (including 1.83.7) cannot share an environment with this release. Use RAE's built-in pricing, run such LiteLLM releases in a separate environment, or select a LiteLLM release whose requirements allow the patched Click version. Optional LiteLLM pricing detection remains available only when its dependencies are compatible; do not downgrade Click to enable it.

[0.13.1] - 2026-08-30

Fixed

  • ModuleNotFoundError: No module named 'httpx' on a fresh install: httpx was imported at module top level in five modules but never declared as a dependency. It had always arrived transitively through mcp, which required httpx up to 1.x. mcp 2.0 switched to httpx2, so new installs resolving MCP 2.x stopped installing httpx and every CLI invocation died at import time, --version included. Environments created before mcp 2.0 kept working, which is why the break only appeared for new users. The package now uses httpx2 (imported as httpx) and declares httpx2>=2.5.0 explicitly. Installs resolving MCP 2.x share this client; environments retaining MCP 1.x also install its legacy httpx dependency. (#122)

[0.13.0] - 2026-07-27

Added

  • client_executed json-stream event: a non-interactive --json/--json-stream run is a single SDK turn end-to-end, so result only fires once the whole turn is over — a caller watching the stream had no earlier way to know a working, live-verified client already existed, even when the agent spent a long tail of that turn on unrelated busywork. --json-stream now emits {"event": "client_executed", "script_path": "..."} as soon as the agent successfully runs the generated client with its own language-specific run command, giving wrapper scripts a signal they can bound cost and time against. Emitted for both engineer and agent; Claude SDK only (OpenCode/Copilot/Cursor have separate streaming handlers). See the scripted usage docs.

Changed

  • Slimmer source distribution: the sdist no longer ships the repo's binary banner/demo assets, the examples/ and design/ trees, or uv.lock — none of which are needed to build, install, or test the package (README images are served from GitHub, not the tarball). The published .tar.gz drops from ~1.0 MB to ~0.2 MB. The wheel is unchanged.

[0.12.0] - 2026-07-22

Changed

  • Playwright is now an optional [manual] extra: agent mode (the default) captures through npx-launched browser tooling (browser MCP servers or the agent-browser CLI) and never imports the Python Playwright package, so playwright and playwright-stealth moved out of the base dependencies into a [manual] optional-dependency group. This keeps the base install (and any runtime that bundles the package) lightweight. Manual capture mode now requires pip install "reverse-api-engineer[manual]" followed by playwright install chromium. Entering manual mode without the extra fails fast — before any run is recorded — with an actionable hint listing both steps, so a base-only install never leaves a phantom manual run in history.

Fixed

  • Manual capture no longer wedges the REPL after a failed start: if a capture failed to launch (for example a mistyped or scheme-less URL), Playwright's sync event loop was left running, so every subsequent prompt died in a loop with "asyncio.run() cannot be called from a running event loop". The browser is now torn down on a failed start, so the error is shown once and the prompt returns to normal.
  • Scheme-less capture URLs are accepted: entering a bare host such as jobs.example.com/x is now treated as https://jobs.example.com/x instead of failing with an invalid-URL error.
  • CLI error text with brackets renders in full: error messages are escaped before Rich renders them, so hints like reverse-api-engineer[manual] are no longer partially swallowed as markup.

[0.11.0] - 2026-07-22

Changed

  • OpenCode setup: RAE now reuses an existing server or downloads/starts opencode-ai@latest through npx without requiring a global OpenCode install, with configurable auto-start, package, and base URL settings plus password inheritance. Fresh configurations default to free opencode/big-pickle. Provider/model pairs are validated before session creation, invalid configurations include current free-model suggestions, and older compatible servers show an upgrade warning. /settings shows a loading spinner, uses a live searchable provider/model picker, saves the pair atomically, and remains open across related changes until Back is selected.
  • Ollama setup: OpenCode mode can discover tool-capable Ollama models, start an installed daemon, and inject provider configuration without modifying the user's opencode.json.

Fixed

  • "Prompt is too long" session deadlock (#93): Claude SDK sessions now run with proactive auto-compaction configured per the Claude Code docs — CLAUDE_CODE_AUTO_COMPACT_WINDOW (capped by the CLI to the model's real context window) plus CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=85 — leaving enough headroom that a max-size HAR read landing near the threshold no longer jumps straight past the hard context limit. Both variables are respected if the user sets them explicitly. When the window is exhausted anyway, the error is now explained (progress is saved on disk; start a new run for the same target to continue) and the follow-up prompt is no longer offered on the dead session, where every further message would fail with the same error.
  • OpenCode permissions: Permission V2 events now reply through OpenCode's current, non-deprecated permission endpoint while retaining compatibility with older servers.
  • OpenCode errors: Known authentication, model configuration, and temporary provider-availability failures now produce one actionable message without the unexpected-error issue prompt.
  • OpenCode TUI prompt echo: Streamed text is now restricted to assistant message IDs, preventing RAE's internal browser and reverse-engineering instructions from appearing as model output.
  • run/list now work for every output language: discover_scripts() previously only found .py files, so reverse-api-engineer run <run_id> failed with "No Python scripts found" for JS/TS/Go/Java/C#/PHP/Ruby/C clients. Discovery now covers all supported extensions (excluding build dirs and the vendored cJSON sources), and the run command dispatches to the right toolchain per language (compile+execute for C), with a clear error when the required tool isn't on PATH.
  • Windows-safe run-command quoting: the Java/C#/PHP/Ruby/C run commands quoted paths with POSIX-only shlex.quote, which cmd.exe/PowerShell parse incorrectly for paths containing spaces. Paths are now quoted per-platform (subprocess.list2cmdline on Windows).

Added

  • Go output language: output_language: "go" is now supported alongside python/javascript/typescript, generating a standard-library-first (net/http, encoding/json) Go program, with the same auth-hardcoding/refresh and bot-detection-fallback guidance as the other languages.
  • Java output language: output_language: "java" is now supported alongside python/javascript/typescript, generating a small Maven project using java.net.http.HttpClient (JDK 11+, no HTTP library dependency) and Gson for JSON, with the same auth-hardcoding/refresh guidance as the other languages.
  • C# output language: output_language: "csharp" is now supported alongside python/javascript/typescript, generating a minimal .NET project using System.Net.Http.HttpClient and System.Text.Json (both part of the .NET 5+ base class library — no NuGet dependency needed), with the same auth-hardcoding/refresh guidance as the other languages.
  • PHP output language: output_language: "php" is now supported alongside python/javascript/typescript, generating a script using the curl and json_encode/json_decode core extensions (ext-curl, ext-json — no Composer dependency needed), with the same auth-hardcoding/refresh guidance as the other languages.
  • Ruby output language: output_language: "ruby" is now supported alongside python/javascript/typescript, generating a script using net/http and json (both part of Ruby's standard library — no gem/Bundler dependency needed), with the same auth-hardcoding/refresh guidance as the other languages.
  • C output language: output_language: "c" is now supported alongside python/javascript/typescript, generating a program using libcurl for HTTP and a vendored cJSON for JSON (C has neither in its standard library), compiled and run as a single {run_command} step, with the same auth-hardcoding/refresh guidance as the other languages.

[0.10.0] - 2026-06-01

Changed

  • Packaging metadata: Refreshed the package description, keywords, and classifiers, bumped Development Status to 4 - Beta, and updated the project URLs (Homepage now points to https://reverseapi.dev, added Documentation and Changelog)
  • Source distribution excludes website/: The marketing site under website/ (Astro/Cloudflare Pages source) is now excluded from the sdist build so it no longer ships inside the PyPI package; the wheel already only packaged src/reverse_api

Added

  • agent_provider: "agent-browser": Shell-driven Vercel agent-browser CLI—RAE prefers an agent-browser binary on PATH, otherwise runs npm install -g <pin> (with a console notice), validates --help, and only then falls back to npx -y <pin> if npm cannot install. Prompts embed the resolved shell prefix plus skills get … / skills list, HAR flows, and optional agent_browser_notes. No bundled browser MCP shim; pin via agent_browser_npx_package / RAE_AGENT_BROWSER_PACKAGE.

Added

  • Cursor SDK support: Added sdk=cursor / --sdk cursor engineering support through a bundled Node bridge around the Cursor TypeScript SDK. Cursor runs use the configured Cursor model (default composer-2), accept MCP server configuration, resume Cursor agents across follow-up turns, and normalize streamed tool output plus token usage into the existing TUI/message-store flow
  • Cursor bridge packaging: Bundled the src/reverse_api/cursor_bridge/ Node package so @cursor/sdk dependencies can be installed on demand when Cursor mode is first used

Fixed

  • Manual REPL model resolution: Follow-up engineering now resolves model settings from the selected SDK, including Cursor, OpenCode, and Copilot
  • Cursor streaming: Buffered Cursor model text before rendering so streamed deltas are shown as coherent blocks and no longer produce stray .. lines or bridge hangs
  • Sync test compatibility: Restored the temporary-file helper used by the existing sync test surface

Removed

  • Chrome extension and native messaging host: The chrome-extension/ workspace and the Python native_host module are removed. The install-host, uninstall-host, and run-host CLI subcommands no longer exist. The extension was an experimental/WIP capture surface that never reached parity with manual and agent modes; deleting it also eliminates a JS dev-tooling supply chain (vite, postcss, picomatch, rollup, prismjs) and the corresponding dependabot churn

Security

  • Drop [pricing] extra (litellm): LiteLLM 1.83.7 patched 3 advisories (1 critical SQLi + 2 high RCE/SSTI) but hard-pins click==8.1.8, which would force a click downgrade for all users. The vulnerable code paths are all in the LiteLLM proxy server, which we do not run — we only used litellm as a library for cost lookups. Removed the optional dependency entirely; pricing.py keeps a graceful import-detect path so users who install litellm independently still get the extended model coverage
  • cryptography >=46.0.7 (was >=46.0.6) — patches a buffer overflow on non-contiguous buffer inputs (medium)
  • pytest >=9.0.3 (was >=8.0.0) — patches vulnerable tmpdir handling (medium, dev only)
  • python-multipart >=0.0.27 (was 0.0.22 transitively via mcp) — patches DoS via large multipart preamble/epilogue (medium)

[0.8.0] - 2026-05-06

Added

  • Agent-friendly CLI surface: First-class non-interactive mode for invocation from other agents and scripts
    • --json on agent and engineer: emit a single structured JSON result (run_id, status, artifacts, usage, error info) instead of streaming TUI output. Auto-suppresses follow-up prompts and TTY-detects at REPL entry
    • --no-interactive on engineer: skip the follow-up loop without forcing JSON output
    • agent --dry-run: pre-flight validation (config, network, MCP availability) without launching a browser
    • agent --headless: run agent mode without a visible browser window (manual mode keeps the human-in-the-loop and does not accept --headless)
    • --json-schema-version: explicit version string (v2) on JSON output so callers can pin their parser
    • Stable usage normalization and error_kind enum: machine-readable error categories and consistent token/cost shapes across SDKs
  • engineer --prompt and engineer --fresh CLI flags: replace the equivalent @id <run_id> [--fresh] <prompt> REPL syntax. With --fresh, --prompt fully replaces the captured run's goal; without --fresh, it layers additional instructions on top
  • --help epilogs: expanded help output with examples for every subcommand

Changed

  • Prompts: extracted to standalone markdown templates under src/reverse_api/prompts/ for easier review and iteration; LLM guidelines tightened
  • Banner: replaced the SVG banner with a new painted-ruins JPG

Removed

  • browser-use and stagehand agent providers: Both retired due to upstream churn. Agent mode now offers only auto (Playwright MCP) and chrome-mcp (Chrome DevTools MCP). Configs with agent_provider set to browser-use or stagehand migrate to auto; the browser_use_model and stagehand_model keys are dropped silently. The [agent] optional-dependency extra is removed
  • Tag system (@record-only, @codegen, @id, @docs, @help): The inline tag syntax inside REPL prompts has been removed. Surviving capabilities have first-class CLI flags / arguments instead:
    • @record-only → manual --no-engineer (already existed)
    • @id <run_id> → engineer <run_id> (positional, already existed)
    • @id <run_id> <prompt> → engineer <run_id> --prompt "..."
    • @id <run_id> --fresh <prompt> → engineer <run_id> --fresh --prompt "..."
    • @codegen and Playwright-action recording: removed entirely with the ActionRecorder and playwright_codegen modules (low usage, replaced by the standard capture+engineer flow)
    • @docs (OpenAPI generation): removed for now; will return as a dedicated docs subcommand in a follow-up
    • @help: superseded by the existing /help slash command in the REPL
  • --reverse-engineer/--no-engineer flag on agent: was parsed but never propagated to the auto-capture path — removed. Agent mode runs an integrated capture + engineering pipeline; use manual --no-engineer for HAR-only recordings

Fixed

  • engineer --json: emits JSON even when RUN_ID is missing (previously fell through to TUI error)
  • Follow-up prompt: suppressed in --json / --no-interactive mode so non-interactive callers never block on stdin

[0.7.1] - 2026-04-06

Fixed

  • Packaging: Exclude .claude/worktrees/, .claude/plans/, and demo GIFs from sdist (29MB → 503KB)

[0.7.0] - 2026-04-06

Added

  • run command: Execute generated scripts directly from the CLI with reverse-api-engineer run <id|name>. Supports fuzzy name matching, --ls to list scripts, --file to pick a specific script, and -- to pass arguments through to the script
  • Shared venv: Scripts run inside a shared venv at ~/.reverse-api/runs/.venv with requests pre-installed. Per-run requirements.txt is installed on top automatically
  • Auto-install missing imports: When a script fails with ModuleNotFoundError, offers to install the missing package and retry

Security

  • litellm >=1.83.0 — fixes critical OIDC auth bypass + high privilege escalation
  • requests >=2.33.0 — fixes insecure temp file reuse
  • aiohttp >=3.13.4 — fixes header injection, SSRF, DoS vulnerabilities
  • pygments >=2.20.0 — fixes ReDoS
  • cryptography >=46.0.6 — fixes incomplete DNS constraint enforcement

[0.6.0] - 2026-04-01

Added

  • Chrome DevTools MCP agent provider (chrome-mcp): Agent mode can drive the browser through Chrome DevTools MCP (--autoConnect, optional --no-usage-statistics) with a dedicated system prompt and SDK wiring alongside the existing Playwright MCP path

Changed

  • Agent provider selection: CLI settings and mode flow distinguish auto (Playwright MCP) from chrome-mcp (Chrome DevTools MCP)
  • Packaging: Source distributions exclude local demo video, packed Chrome extension zip, store-asset screenshots, and per-machine .claude/settings.local.json files so PyPI artifacts stay small (wheel unchanged: src/reverse_api only)

Fixed

  • Tool result blocks (engineer / agent streaming): When content is empty, fall back to result or output on tool-result blocks so alternate SDK shapes still surface output in the UI and message store
  • Process title: Set when the CLI module loads so the terminal shows reverse-api earlier in startup
  • Follow-up prompt: Flush local sync before reading follow-up input so state is up to date
  • Tests: ClaudeAutoEngineer analyze tests now use real claude_agent_sdk message types (so isinstance checks in the streaming loop match) and stub _prompt_follow_up to avoid the interactive follow-up path

Documentation

  • README: Agent mode demo GIF

[0.5.0] - 2026-03-17

Added

  • Follow-up chat: After a run completes, type follow-up messages to iterate in the same session without creating new run IDs or folders. Press Enter to finish
  • Abort run (Ctrl+C): Gracefully cancel a running agent/engineer session and return to the REPL instead of exiting the app
  • AskUserQuestion free mode: All select/checkbox prompts now include "Other (type your answer)" so users can always provide free-text input. Updated agent prompt to document free-form, multi-select, and multi-question capabilities
  • Random task suggestions (Ctrl+R): Press Ctrl+R in agent mode to fill the prompt with a random curated task idea. Press again to cycle

Fixed

  • Usage tracking across follow-ups: Token counts now accumulate across all turns in a session instead of being overwritten by the last turn
  • Agent mode follow-up: Auto engineer (agent mode) was using a copy-pasted streaming loop without follow-up support; now reuses the shared conversation loop

[0.4.5] - 2026-03-15

Fixed

  • Stream closed errors (#51): Reverted from query() to ClaudeSDKClient which maintains a persistent bidirectional connection, eliminating "Error in hook callback hook_0: Stream closed" errors. The original AUQ fix (v0.4.3) unnecessarily switched APIs — only the can_use_tool callback signature needed updating
  • CLAUDECODE env var leak: Clear inherited CLAUDECODE env var from CLI subprocess to prevent nested session interference when running inside Claude Code
  • CLI stderr noise: Filter minified JS stack traces into a single clean error line (use DEBUG=1 for full output)

Changed

  • claude-agent-sdk: Bumped minimum version to 0.1.48
  • Agent mode: No longer prompts for URL (agent navigates autonomously)
  • Header UI: Version and task labels now use mode-specific colors (agent=coral, engineer=blue, collector=gold)

[0.4.4] - 2026-03-15

Fixed

  • Stream closed errors (#51): Partial fix using query() with env var workarounds (superseded by v0.4.5)

[0.4.3] - 2026-03-12

Fixed

  • AskUserQuestion interactive prompt: Fixed interactive questionary UI not rendering in engineer and agent modes. Switched from ClaudeSDKClient to query() function, as ClaudeSDKClient silently auto-approves AskUserQuestion without triggering the can_use_tool callback

[0.4.2] - 2026-03-12

Fixed

  • Engineer client resolution: Prefer the active engineer client recorded in session history over config defaults, preventing iterative edits from switching to the wrong file when multiple language clients exist

[0.4.1] - 2026-03-12

Added

  • show CLI subcommand: View details of a specific run by ID
  • Chrome Extension improvements: Multi-session support, codegen recording, side panel UX overhaul
    • Session isolation fixes to prevent capture data bleeding across sessions
    • Message persistence across panel reloads and session switches
    • Traffic list with expandable request details and clear button
    • Restricted page detection with user feedback
    • ANSI escape code stripping in terminal output
    • PrismLight language fallback for unregistered languages
    • Bundle size reduced from 1112KB to 436KB (removed shiki, lucide-react, ansi-to-react, @base-ui/react)
  • Mintlify API client example

Changed

  • Background update check: Version check now runs in a background thread to avoid blocking CLI startup

Fixed

  • Engineer language preservation: Iterative edits now preserve the original output language instead of resetting
  • Native host Gatekeeper handling: Scoped xattr to the actual claude-code package directory
  • Codegen selector escaping: Dynamic attribute values now escaped with CSS.escape() to prevent broken selectors
  • Chrome Extension manifest: Removed unused scripting permission; pinned devicon CDN to v2.16.0

[0.4.0] - 2026-03-10

Added

  • list CLI subcommand: Non-interactive command to list generated scripts and runs
    • Rich table output (compact and full modes) and JSON output (--json)
    • Filter by mode (--mode), model (--model), prompt search (--search), and limit (--limit)
    • Shows script directory paths, file counts, and local path detection

[0.3.3] - 2026-03-10

Added

  • GitHub Copilot SDK: Third SDK option for reverse engineering using GitHub Copilot
    • New CopilotEngineer for HAR analysis via Copilot subscription (cost: $0)
    • Auto mode support with MCP browser integration via Copilot SDK
    • Interactive AskUserQuestion tool support for Copilot sessions
    • Permission handler and tool use hooks for agent visibility
    • Install with: pip install 'reverse-api-engineer[copilot]'
  • Comprehensive Test Suite: 593 tests achieving 97.4% code coverage

Changed

  • Claude Model Updates: Updated from Claude 4.5 to Claude 4.6 (Opus and Sonnet)
  • Sync Filtering: Unified sync filtering logic with relative paths; excludes node_modules
  • TUI Improvements: Reduced thinking truncation for better agent visibility

Fixed

  • Python 3.11 Compatibility: Fixed multi-line f-string expressions that required Python 3.12+
  • Thread Safety: Used loop.call_soon_threadsafe for Copilot SDK event callbacks
  • Session Timeouts: Added 10-minute timeout protection for Copilot session completion
  • Resource Cleanup: Ensured CopilotClient is always stopped via try/finally
  • HAR Recording: Changed HAR recording content mode to embed and optimized file saving

[0.3.2] - 2026-01-15

Added

  • Update Notifications: CLI now checks for newer versions on startup and displays update prompt

Changed

  • README Improvements: Redesigned badge styling with flat-square style and centered layout
  • Banner Redesign: Modernized SVG banner with centered content and improved visual hierarchy
  • Font Compatibility: Fixed banner fonts for better GitHub rendering support

Fixed

  • Banner Display: Corrected package name display and font rendering in GitHub

[0.3.1] - 2026-01-15

Added

  • Chrome Extension (WIP): Beta support for capturing browser traffic via Chrome extension
    • Alternative to Playwright browser for HAR capture
    • Works with existing browser sessions
    • Note: This feature is work in progress and may have limitations

Fixed

  • OpenCode Server: Fixed error formatting and connection handling
  • Process Title: Added process title for better identification in system monitors
  • Auth Error Handling: Improved handling of authentication failures

[0.3.0] - 2026-01-10

Added

  • Collector Mode: New AI-powered web data collection mode using Claude Agent SDK
    • Natural language prompts to collect structured data from any website
    • Automatic export to JSON and CSV formats
    • Generates README with collection metadata and schema
    • Uses WebFetch, WebSearch, and file tools for autonomous collection
  • Playwright Codegen: Generate automation scripts from recorded browser actions
    • Captures clicks, fills, key presses, and navigations
    • Produces stealth-enabled Playwright scripts with proper escaping
    • Deduplicates redundant fill actions and navigations
  • @docs Tag: Generate OpenAPI specifications from HAR files
    • Standalone usage: @docs to generate from latest run
    • With run ID: @docs run_id to generate from specific run
  • @record-only Tag: Record HAR files without reverse engineering step
  • AskUserQuestion Tool: Interactive prompts during engineering sessions
  • JS/TS Client Generation: Support for generating JavaScript/TypeScript API clients

Changed

  • Improved HAR Filtering: Better path-based filtering for skip patterns
  • Enhanced Price Computation: Fixed pricing for OpenCode provider
  • Centralized Run Resolution: Refactored latest run parsing logic

Fixed

  • Selector Escaping: Fixed attribute selector escaping for special characters in name, data-testid, aria-label, and placeholder values
  • Null Checks in Codegen: Added validation for action.selector and action.value to prevent crashes
  • CSV Export: Fixed DictWriter error when items have inconsistent keys
  • HAR Validation: Validate HAR file exists when using @docs tag
  • Path Normalization: Fixed path handling in various utilities

[0.2.10] - 2026-01-03

Added

  • Claude Code Plugin: Official plugin for seamless integration with Claude Code CLI
    • Three operation modes: manual browser capture, autonomous agent browsing, and re-engineering from HAR files
    • Comprehensive skill system with progressive disclosure of reverse engineering techniques
    • Slash commands for quick access: /agent, /engineer, /manual
    • Reference documentation for HAR analysis and authentication patterns
    • API client templates for common patterns
  • Example API Clients: Added production-ready examples for major platforms
    • Apple Jobs API client with field extraction utilities
    • Ashby Jobs API client with comprehensive endpoint coverage
    • Ikea API client for product search and catalog browsing
    • Uber Careers API client with pagination support
  • Engineer Tagging System: Enhanced metadata tracking for generated API clients
    • Automatic tagging of runs with descriptive identifiers
    • Improved organization and searchability of reverse-engineered APIs

Changed

  • Enhanced Auto Mode: Improved MCP browser integration with better error handling
  • Better Sync Fallback: More robust file synchronization with fallback mechanisms
  • Code Quality: Comprehensive formatting and linting improvements across codebase

Fixed

  • Agent Mode Screenshots: Reduced unnecessary screenshot captures in agent mode
  • Sync Error Handling: Fixed sync fallback when primary sync method fails
  • Path Handling: Corrected CLAUDE.md documentation paths
  • Import Errors: Fixed missing imports in various modules

[0.2.9] - 2025-12-30

Added

  • Real-time File Sync: Watch and automatically sync generated scripts to local directory
    • Debounced file watching with configurable delay (default 500ms)
    • Visual feedback for sync operations in terminal UI
    • Prevents overwriting existing directories by appending counter suffix
  • MCP Browser Integration: Native integration with rae-playwright-mcp for auto mode
    • Seamless browser automation via Model Context Protocol
    • Works with both Claude SDK and OpenCode SDK
    • Combines browser control and real-time reverse engineering in single workflow
  • CLAUDE.md Autogeneration: Automatic generation of project documentation for Claude Code

Changed

  • Enhanced Settings Management: Improved settings configuration and UI
  • Better Sync Error Handling: Improved error handling and resource cleanup for sync operations

Fixed

  • Sync Directory Overwrite: Fixed issue where sync would overwrite existing directories
  • Sync Resource Leaks: Fixed memory leaks when sync errors occurred
  • UI Improvements: Various UI fixes and enhancements

[0.2.8] - 2025-12-28

Added

  • 3-Tier Pricing Fallback System: Automatic pricing lookup for 100+ LLM models
    • Local pricing for common models (highest priority)
    • Optional LiteLLM integration for extended coverage (install with pip install 'reverse-api-engineer[pricing]')
    • Default fallback to Claude Sonnet 4.5 pricing
  • New Model Pricing: Added pricing for Gemini 3 and Claude thinking series models

Changed

  • Enhanced OpenCode Prompts: Improved prompt handling for code generation
  • Better Folder Naming: Folder name generation with OpenCode SDK
  • Antigravity Documentation: Added comprehensive documentation for free models via Antigravity

Fixed

  • Model name mismatch in pricing lookups
  • Pricing computation for extended thinking models

[0.2.7] - 2025-12-27

Changed

  • Version management: Implemented single source of truth for versioning
    • Version now defined only in pyproject.toml
    • __init__.py reads version dynamically using importlib.metadata
    • Eliminates need to manually update version in multiple files
    • Added RELEASING.md with release process documentation

[0.2.6] - 2025-12-27

Fixed

  • Version flag: Updated __version__ to 0.2.6 to ensure --version displays correctly
  • OpenCodeEngineer initialization: Refactored to properly pop specific kwargs (opencode_provider and opencode_model) before passing to parent class
    • Ensures only relevant arguments are sent to BaseEngineer
    • Improves initialization logic clarity and prevents unintended argument passing

Changed

  • README improvements: Added table of contents and removed repetitive sections for better readability

[0.2.5] - 2025-12-27

Fixed

  • Initial release attempt (superseded by 0.2.6 due to missing version flag update)

[0.2.4] - 2025-12-27

Fixed

  • Version string: Fixed --version flag to correctly display 0.2.4 instead of outdated 0.2.0
    • Previous release (0.2.3) was built with stale bytecode cache
    • Added clean build script (scripts/clean_build.sh) to prevent future stale builds

[0.2.3] - 2025-12-27

Changed

  • Version display: Fix hardcoded version display
  • Logs display: Remove agent logs, claude agent sdk logs
  • Browser-use installation: Better instructions on how to install bu for agent mode

[0.2.2] - 2025-12-26

Changed

  • Better HAR Recording: Improved HAR file recording and capture functionality

[0.2.1] - 2025-12-26

Added

  • Stagehand Agent Support: Added Stagehand as an alternative agent provider alongside browser-use
    • Supports OpenAI Computer Use models (e.g., computer-use-preview-2025-03-11)
    • Supports Anthropic Computer Use models (e.g., claude-sonnet-4-5-20250929, claude-haiku-4-5-20251001, claude-opus-4-5-20251101)
  • Separate Model Configurations: Enhanced settings system with independent model configurations
    • claude_code_model: Model for Claude SDK (renamed from model)
    • opencode_provider: Provider for OpenCode SDK (e.g., "anthropic", "openai", "google")
    • opencode_model: Model for OpenCode SDK
    • browser_use_model: Model for browser-use agent provider
    • stagehand_model: Model for stagehand agent provider

Changed

  • Improved Settings Management: Separated model configurations for different SDKs and agent providers
    • Each SDK and agent provider now has its own independent model setting
    • Settings menu updated with clearer options for each component
  • Better Configuration Isolation: OpenCode model settings no longer interfere with Claude SDK settings
  • Backward Compatibility: Automatic migration of old config files to new structure

Fixed

  • Fixed issue where OpenCode model settings were being overridden by Claude SDK model settings
  • Fixed model configuration conflicts between different SDKs and agent providers

[0.2.0] - 2025-12-25

Added

  • OpenCode SDK Support: Native integration with OpenCode SDK for more flexibility in reverse engineering workflows
  • Agent Mode: Fully automated browser interaction using AI agents (browser-use) with support for multiple LLM providers
    • Browser-Use LLM (default)
    • OpenAI models (gpt-4, gpt-3.5-turbo, etc.)
    • Google models (gemini-pro, gemini-1.5-pro, etc.)
  • Multi-Provider Agent Support: Configure agent models via settings with automatic API key detection

Changed

  • Improved UX with better CLI interactions and mode cycling
  • Enhanced settings management for model, agent model, SDK, and output directory configuration
  • Better error handling and user feedback throughout the application

[0.1.0] - 2025-12-22

Added

  • Initial release
  • Browser automation with Playwright and stealth mode
  • HAR recording and capture
  • AI-powered API client generation using Claude
  • Interactive CLI with manual and engineer modes
  • Session history and cost tracking
  • Production-ready code generation with type hints and documentation