Prompt file imported from shikharsumantech14/parallax (.claude/commands/pipeline-research.md). Fill in {{arguments}} before use. Copyright stays with the author.

/pipeline-research

Step 2 of the Parallax editorial pipeline. Deep-researches a chosen candidate and produces a verified dossier for the drafter.

Usage

/pipeline-research politics
/pipeline-research space

Argument is one of: politics, space, earth, tech, travel, sports.

Prerequisite: You must have already run /pipeline-discover <category> and picked a candidate by changing its status: open → status: chosen in the candidates file.

What this does

1. Validates the category argument
2. Finds the most recent candidates file for the category
3. Confirms exactly one candidate has status: chosen
4. Spawns the researcher subagent with the category and chosen candidate as context
5. The agent verifies all facts, finds primary sources and key quotes, proposes the issue structure, and writes the dossier
6. Returns the dossier path and a summary of what was found

What you do next

1. Open research/<category>/<YYYY-MM-DD>-<slug>-dossier.md
2. Read through the structural argument, verified facts, and suggested structure
3. Check any items marked [UNVERIFIED] — decide if they need resolving before drafting or can be dropped
4. If the dossier looks solid, run /pipeline-draft <category> (Phase 3 — coming soon)
5. If the dossier has gaps, you can ask the researcher to go deeper on a specific section before moving to draft

Cost

Each run uses:

• ~10-20 WebSearch calls
• ~8-15 WebFetch calls (one per primary source + cross-checks)
• One Sonnet pass for synthesis and structure

Approx ₹30-70 per run on Anthropic API direct, or ~5-10% of a Claude Pro 5-hour limit window.

Instructions to Claude

The user has invoked /pipeline-research with argument: {{arguments}}

1. Validate that {{arguments}} is one of: politics, space, earth, tech, travel, sports. If not, print the valid options and stop.

2. Verify research/_sources/{{arguments}}.md exists. If not, stop.

3. Glob research/{{arguments}}/*-candidates.md. Read the most recent file. Check for a candidate with status: chosen.

   • If none found: tell the user to run /pipeline-discover {{arguments}} first and pick a candidate.
   • If more than one is chosen: tell the user to unset the extras and leave exactly one status: chosen.

4. Spawn the researcher subagent with this prompt:

   Run research for category {{arguments}}. The candidates file is at research/{{arguments}}/<most-recent-candidates-file>. Find the candidate with status: chosen, deeply research it using the source allowlist at research/_sources/{{arguments}}.md, and write the dossier to research/{{arguments}}/<today-YYYY-MM-DD>-<slug>-dossier.md. Follow the template at research/_templates/dossier.md exactly. Return a one-paragraph summary with: dossier file path, the structural argument in one sentence, 3 strongest verified facts, and any [UNVERIFIED] items to flag for the editor.

5. When the subagent finishes, relay its summary to the user. Do not add editorial commentary — the editor reviews the dossier directly.