LegoFlow

BlocksTerminal Curator

Getting Started

Requirements

Use Python 3.12, uv, and a Docker-capable CPU host. Generation also needs an OpenAI-compatible LLM endpoint/model/key; live collection needs Stack Exchange access. Local candidate imports need no generation key. Setup prepares the pinned repositories and isolated environments.

Install the plugin

From the repository root, choose your agent:

# Claude Code; then reload plugins or restart the session.
claude plugin marketplace add ./blocks/terminal-curator/.claude/plugins
claude plugin install terminal-curator@terminal-curator-block

# Codex; then restart the session.
codex plugin marketplace add .
codex plugin add terminal-curator@legoflow

Configure a small batch

Edit blocks/terminal-curator/config.yaml. Choose source.type: stackoverflow, json, or tasks; set source.path for the latter two. Start with max_questions: 1, generation.workers: 1 and validation.n_concurrent: 1.

For generation, set llm_api.api_base_url and llm_api.model, leave its tracked key empty, and supply OPENAI_API_KEY through the execution environment. See Configuration Guide for all source and profile options.

Check and run

ActionClaude CodeCodex
Prepare/terminal-curator:setup$terminal-curator:setup
Preflight/terminal-curator:check$terminal-curator:check
Run/terminal-curator:run$terminal-curator:run
Monitor/terminal-curator:dashboard$terminal-curator:dashboard

The agent reports the resolved source, host, model and budget, then performs an authorized run. Script equivalents, from blocks/terminal-curator/:

bash scripts/setup.sh
bash scripts/dryrun.sh
bash scripts/start.sh
artifacts/env/terminal-curator/bin/python dashboard/progress_monitor.py --html

Pass the same --config <file> to each command when using a private full profile. Run checks and containers on the selected Docker host; synchronize code, inputs and profiles before remote execution through SSH and tmux. Carry resource variables into the actual execution window: new tmux windows inherit the session/server environment, which may differ from the current shell. Use session-scoped settings or a private environment file for credentials.

Inspect the result

Check the run report, archive and artifacts/tasks/verifiable_tasks.txt. The run skill completes validation and semantic review, including fresh trials after corrections, before handing tasks downstream. Then follow Downstream Handoff to select the publication in Tracer or Evaluator.

On this page