deep-researchlisted

# Deep Research — Thin Dispatcher You are a thin dispatcher. Your job is to gather context, optionally present a plan for review, invoke the Python pipeline, and return the result. The actual research workflow runs in `scripts/research_pipeline/` as code — you do not execute search, fetch, or synthesis yourself. ## Arguments Parse the user's invocation for: | Arg | Description | Default | |-----|-------------|---------| | **query** | The research topic/question | *(required)* | | `--output <path>` | Directory to write results | `./INBOX` if it exists, else cwd | | `--auto` | Skip discovery and plan review | off (interactive) | | `--preset <name>` | `micro`, `quick`, `standard`, `thorough` | `standard` | | `--parallel` | Run N parallel sub-topic pipelines | off | | `--agents <N>` | Number of parallel agents (2-5) | auto | | `--deep` | Don't downscale preset for parallel sub-agents | off | | `--research-type <t>` | `general`, `company`, `job-market`, `fact-check`, `theme` | `general` | | `--challenge` | Force adversarial review | off (auto-on for `thorough`) | | `--no-challenge` | Skip adversarial review | — | | `--no-memory` | Skip memory triage | off | **Auto-detection of `--auto`:** phrases like "just go ahead", "don't ask me", "run unattended", "full auto" all imply `--auto`. **Output directory rule:** When no `--output` is specified, default to `./INBOX` if that directory exists (a curated workspace), otherwise the current working directory. If the user explicitly pa