The connector hooks into Cline’s event system and captures structured data from each task. This page documents what events are captured and how they map to Infinium traces.

Hook Events

The connector listens for these Cline hook events:

Event	When It Fires	What’s Captured
TaskStart	A new task begins	Task description, model (provider + slug), workspace root, Cline version
TaskResume	A previous task is resumed	Same as TaskStart
UserPromptSubmit	User sends a follow-up prompt in the task	Prompt text
PreToolUse	Before a tool executes	Tool name, parameters
PostToolUse	After a tool completes	Tool result, success flag, durationMs
PreCompact	Cline is about to compact the conversation	Conversation length, estimated tokens
TaskComplete	Task finished successfully	Final task text — triggers trace send
TaskCancel	Task was cancelled	Triggers trace send

Nested JSON Payload

Unlike most other Claude-family tools, Cline sends nested JSON. Each hook event carries a sub-object keyed by the hook name:

{
  "hookName": "PostToolUse",
  "taskId": "cline-t-abc123",
  "timestamp": "2026-04-18T10:00:05Z",
  "clineVersion": "3.36",
  "workspaceRoots": ["/home/user/project"],
  "userId": "u-1",
  "model": {"provider": "anthropic", "slug": "claude-sonnet-4"},
  "postToolUse": {
    "tool": "execute_command",
    "parameters": {"command": "npm test"},
    "result": "All 42 tests passed",
    "success": true,
    "durationMs": 3200
  }
}

The connector normalizes hookName → hook_event_name and taskId → session_id internally so downstream logic matches the other connectors.

Stdout Response

Cline requires every hook invocation to write a JSON response on stdout. The connector always writes:

{"cancel": false, "contextModification": null, "errorMessage": null}

If this response is missing or malformed, Cline will hang. The connector guarantees a response is always written, even on error.

How Events Become Traces

A full task lifecycle looks like this:

Task:
  TaskStart         → "Fix the login validation bug"
  UserPromptSubmit  → "Verify with: npm test"
  PreToolUse        → execute_command("npm test")
  PostToolUse       → {result: "passed", success: true, durationMs: 3200}
  TaskComplete      → "Fix the login validation bug"
                       ↓
                    Trace sent to Infinium

The trace is sent on TaskComplete (or TaskCancel). One trace per task.

Trace Structure

Each trace sent to Infinium contains:

Top-Level Fields

Field	Source	Example
name	Task text (truncated to 30 chars)	"Fix the login bug"
description	Task text (full)	"Fix the login validation bug"
duration	Task timing	15.2 (seconds)
input_summary	First prompt / task text	"Fix the login validation bug"
output_summary	Last assistant response (if present)	"Fixed and tests pass"

Execution Steps

Each trace includes an ordered list of steps:

Step Type	Action	Fields
User prompt	user_prompt	description, input_preview, output_preview
Tool call	tool_use	description (e.g. "Tool: execute_command"), input_preview, output_preview, duration_ms
Tool failure	tool_use with error	error.error_type = "ToolError", error.message, error.recoverable = true

Environment Context

Every trace includes environment metadata:

Tag	Value
model	Combined provider/slug (e.g. anthropic/claude-sonnet-4)
model_provider	Just the provider (anthropic, openai, etc.)
session_id	Cline task identifier
working_directory	Workspace root
cline_version	Cline extension version
framework	"cline"

LLM Usage

On flush, the connector reads Cline’s ui_messages.json (in VS Code’s global storage for the saoudrizwan.claude-dev extension) and attaches cumulative token usage to the trace:

Field	Source
prompt_tokens	Sum of tokensIn across api_req_started entries
completion_tokens	Sum of tokensOut
api_calls_count	Number of api_req_started entries
provider	From model.provider
model	From model.slug
cache_reads, cache_writes	Surfaced in custom_tags (no dedicated SDK field)
total_cost	Surfaced in custom_tags

Tool Names

The connector understands Cline’s common tool set and normalizes their display:

execute_command, shell, read_file, readFile, write_to_file, writeFile, replace_in_file, editFile, search_files, grep, list_files, browser_action.

Error Tracking

PostToolUse events with success: false are captured as structured errors:

Field	Description
error_type	"ToolError"
message	Cline’s result text (truncated)
recoverable	true

Data Truncation

To keep payloads manageable, the connector truncates:

• Prompt text: First 200 characters (configurable via CONNECTOR_MAX_INPUT_PREVIEW)
• Tool output preview: First 300 characters (configurable via CONNECTOR_MAX_OUTPUT_PREVIEW)
• Payload size: Progressive truncation if the total payload exceeds ~950KB

Full data is never sent — only summaries that give Maestro enough context to evaluate quality.