A multi-agent research skill for Claude. Given any topic, Parallax assembles a team of analyst agents — each with a distinct identity, agenda, and research mission — then synthesises their findings into a single structured report.
| Report | Topic | Agents |
|---|---|---|
| palantir.md | Palantir (PLTR) stock outlook | 7 |
| iran-us-truce.md | Iran–US truce likelihood before end of May 2026 | 7 |
| ai-threat-to-saas.md | Is the AI threat to SaaS significant in the short, medium, or long term? | 7 |
| politically-driven-fed-rate-cut.md | Politically-driven Fed rate cuts against market consensus | 7 |
This is a work in progress, please note:
- Works as a skill and can be run from any Claude front end: Chat, Cowork, Code, and API.
- Project context, other Skills and MCP will be used if found to be appropriate for search otherwise websearch is used.
- A large number of views and websearches can result in a high token cost, but - possibly due to the structure provided by the agent framework - good results can be obtained with less capable models - I'm using Sonnet 4.6.
- An executive synthesis with convergent findings, key disagreements, and open questions
- A comparison table (when the topic fits)
- Per-agent deep-dives with inline citations
- Downloadable report in Word, PDF, or Markdown
- You give it a topic
- It proposes a roster of analyst agents (e.g. short-seller, regulator, affected community member)
- You approve or edit the roster — add agents, remove them, rename them
- Agents are deployed to research the topic from their own viewpoint, using web search, MCP tools, loaded skills, and any context you've shared
- Findings are synthesised into a report
Agents can run sequentially (with live progress notes) or in parallel (all at once, much faster). Note parallelism doesn't work in the Claude Chat interface
Drop SKILL.md into your Claude skills directory:
| Environment | Skills directory |
|---|---|
| Claude Code | ~/.claude/skills/parallax/SKILL.md |
| Claude Cowork | Via the Claude Desktop skills panel |
Then restart your Claude environment.
/parallax "was Brexit a success"
/parallax "Palantir stock outlook"
/parallax "should my city ban cars from the centre"
Or just describe what you want — Parallax also triggers on natural language like "give me a multi-perspective analysis of X" or "steelman both sides of Y".
After the roster is proposed, you can edit it freely before research starts:
add a labour union organiser
remove 3
drop 2 and 5
add Economist and remove the regulator
| Mode | Speed | Mid-run control | Environments |
|---|---|---|---|
| Sequential | Slower | ✅ Say "more depth" | All |
| Parallel | Much faster | ❌ | Claude Code, Claude Cowork |
After the report is printed, reply with the format you want:
"word" → .docx (requires pandoc, python-docx, or Google Drive MCP)
"pdf" → .pdf (requires pandoc)
"md" → .md (no dependencies)
- A Claude environment that supports skills (Claude Code or Claude Cowork)
- At least one research tool available (web search, MCP server, or relevant loaded skills)
- Parallel mode additionally requires subagent support (Claude Code or Claude Cowork)
MIT — see LICENSE.