Update the dbt_project.yml file with task-specific Snowflake credentials.

Update the profiles.yml file with task-specific Snowflake credentials.

Upload entire run directory to S3 bucket.

Add files to exclusion list.

Debug breakpoint that cleanly halts task execution. Args: message: Optional message to display before halting logger: Option

Load catalog on demand.

Just run all of the checks.

Print check results.

Timeout for cleanup operations in seconds.

Copy files or directories to a running docker container.

Default timeout for agent task execution in seconds.

Default timeout for test execution in seconds.

Discover DuckDB files in the configured source directory.

Migrate DuckDB databases to Snowflake.

Extract deduplicated tool names from Claude Code agent logs. Filters out generic tools (Bash, Edit, Glob, etc.) and expands

List of path patterns to exclude from file diffing.

Find all file_diff_log.txt files in the experiment directory.

Format the Claude Code agent's log file into a human-readable string. Also generates an HTML transcript at log_path.parent / "transc

Format the parsed turns into a readable text string.

Create from dictionary.

Create snapshot from dictionary.

Generate an HTML transcript from the log file. Override this method in subclasses to provide HTML transcript generation. The

Get a copy of a database file for use outside of containers. Note: For container use, it's more efficient to use find_database_file(

Get information about a specific database.

Get default timeouts when no task context is available.

Get the number of diffs created.

Get the most recent snapshot.

Get an existing tmux session by name.

Check if this task has a variant matching the given db_type and project_type.

Launch an interactive shell into a task environment. This command sets up the task environment exactly like the harness would, then drop

Launch an interactive shell into a task environment. This command sets up the task environment exactly like the harness would, then drop

List available tasks with their details and prompts.

List all available shared databases.

Load FileDiffHandler from an optimized diff log file.

ADE-bench: Analytics and Data Engineering Benchmark

Parse Claude agent response to extract metrics. Returns a dictionary with the following keys: - runtime_ms: The higher of du

Parse Macro agent response to extract metrics. Returns a dictionary with the following keys: - runtime_ms: Currently set to

Parse Gemini agent response to extract metrics. Gemini outputs JSON at the end with format: {"response": "...", "stats": {"m

Parse Codex agent response to extract metrics. Codex outputs JSON lines with format: {"type":"turn.completed","usage":{"inpu

Calculate pass@k metrics for different k values. For each k, returns the percentage of tasks where at least one attempt was

Register a new database in the shared pool. Args: db_path: Path to the database file description: Descriptio

Remove a database from the shared pool.

Run ADE-bench with specified tasks and configuration. Example: ab run airbnb001 --db duckdb --project-type dbt --agent sage

Path to the run_sql.py utility script.

Path to the run_sql.sh utility script.