Skip to content

The agent that turns websites into APIs!

chore: update branch

kalil0321d95a8715a673
diff --cc CLAUDE.md
index 0097e96,82f2f6b..11143fa
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@@ -29,8 -29,8 +29,10 @@@ Both inherit from `BaseEngineer` (abstr
### Data Flow
1. Browser captures HAR → saved to `~/.reverse-api/runs/har/{run_id}`
++1. Browser captures HAR → saved to `~/.reverse-api/runs/har/{run_id}`
2. Engineer analyzes HAR with LLM → generates Python scripts
3. Scripts saved to:
++ - `~/.reverse-api/runs/scripts/{run_id}` (permanent)
- `~/.reverse-api/runs/scripts/{run_id}` (permanent)
- `./scripts/{descriptive_name}/` (local copy)
diff --cc src/reverse_api/auto_engineer.py
index 3e699a7,bd9dab1..67663e5
--- a/src/reverse_api/auto_engineer.py
+++ b/src/reverse_api/auto_engineer.py
@@@ -74,17 -75,29 +74,30 @@@ class ClaudeAutoEngineer(ClaudeEngineer
Follow this workflow step-by-step:
### Phase 1: BROWSE
- Use browser MCP tools to accomplish the mission goal:
- - `browser_navigate(url: str)` - Navigate to a URL
- - `browser_click(selector: str)` - Click an element
- - `browser_type(selector: str, text: str)` - Type text into input
- - `browser_wait_for_selector(selector: str)` - Wait for element
- - `browser_screenshot()` - Take screenshot for context
+ Use browser MCP tools to accomplish the mission goal, here are some of the tools you can use.
+ This list is not exhaustive, but it's a good starting point:
+ - `browser_navigate` - Navigate to a URL
+ - `browser_click` - Click an element
+ - `browser_scroll` - Scroll the page
+ - `browser_close` - Close the browser
+ - `browser_evaluate` - Evaluate JavaScript code
+ - `browser_press_key` - Press a key
+ - `browser_run_code` - Run Playwright code
+ - `browser_type` - Type text into input
+ - `browser_wait_for` - Wait for text to appear or disappear or a specified time to pass
+ - `browser_snapshot` - Get accessibility tree (useful alternative to screenshots for understanding page structure)
+ - `browser_take_screenshot` - Take screenshot for context (IMPORTANT: Prefer element-specific screenshots over full-page to avoid size limits)
- And other browser MCP tools available
+ **Screenshot Guidelines:**
+ - Screenshots have a 1MB size limit - avoid full-page screenshots when possible
+ - Prefer taking element-specific screenshots using CSS selectors
+ - If you need context, take multiple smaller screenshots of key areas
+ - Use `browser_snapshot()` when you need page structure information without visual details
+
### Phase 2: MONITOR
-While browsing, periodically call `browser_network_requests` to monitor API traffic in real-time if needed, keep in mind that you will also have access to the full network traffic when closing the browser:
+While browsing, periodically call `browser_network_requests()` to monitor API traffic in real-time.
+Keep in mind that you will also have access to the full network traffic when closing the browser:
- Analyze requests and responses
- Identify authentication patterns (cookies, tokens, headers)
- Note API endpoints, methods, parameters
diff --cc src/reverse_api/browser.py
index 69238c6,4b610f7..cf05725
--- a/src/reverse_api/browser.py
+++ b/src/reverse_api/browser.py
@@@ -669,6 -680,6 +669,7 @@@ class AgentBrowser
# Set before importing to ensure it takes effect
os.environ["BROWSER_USE_LOGGING_LEVEL"] = "WARNING"
++ os.environ["BROWSER_USE_LOGGING_LEVEL"] = "WARNING"
# Import browser-use after setting environment variable
try:
diff --cc src/reverse_api/opencode_engineer.py
index 5ffbfd3,d1731fb..7fe98cd
--- a/src/reverse_api/opencode_engineer.py
+++ b/src/reverse_api/opencode_engineer.py
@@@ -44,6 -44,6 +44,7 @@@ class OpenCodeEngineer(BaseEngineer)
self.opencode_provider = kwargs.pop("opencode_provider", "anthropic")
self.opencode_model = kwargs.pop("opencode_model", "claude-sonnet-4-5")
++
super().__init__(*args, **kwargs)
# Override UI with OpenCode-specific version
@@@ -426,19 -461,19 +444,34 @@@
else:
cost = api_cost
+ self.usage_metadata["input_tokens"] = (
+ self.usage_metadata.get("input_tokens", 0) + input_tokens
+ )
+ self.usage_metadata["output_tokens"] = (
+ self.usage_metadata.get("output_tokens", 0) + output_tokens
+ )
+ self.usage_metadata["reasoning_tokens"] = (
+ self.usage_metadata.get("reasoning_tokens", 0) + reasoning_tokens
+ )
+ self.usage_metadata["cache_read_tokens"] = (
+ self.usage_metadata.get("cache_read_tokens", 0) + cache_read
+ )
+ self.usage_metadata["cache_creation_tokens"] = (
++ self.usage_metadata.get("cache_creation_tokens", 0) + cache_write
++ )
+ self.usage_metadata["input_tokens"] = (
+ self.usage_metadata.get("input_tokens", 0) + input_tokens
+ )
+ self.usage_metadata["output_tokens"] = (
+ self.usage_metadata.get("output_tokens", 0) + output_tokens
+ )
+ self.usage_metadata["reasoning_tokens"] = (
+ self.usage_metadata.get("reasoning_tokens", 0) + reasoning_tokens
+ )
+ self.usage_metadata["cache_read_tokens"] = (
+ self.usage_metadata.get("cache_read_tokens", 0) + cache_read
+ )
+ self.usage_metadata["cache_creation_tokens"] = (
self.usage_metadata.get("cache_creation_tokens", 0) + cache_write
)
self.usage_metadata["cost"] = self.usage_metadata.get("cost", 0) + cost