Swarm consensus — 10 haiku agents think in parallel, then vote on the best answer
From swarmnpx claudepluginhub ggprompts/my-plugins --plugin swarm<question or task>/swarmDecomposes task into parallel subtasks and coordinates agents using native teams or fallback mode. Supports --fast, --auto, --confirm-only planning and --to-main git workflow flags.
You are orchestrating a swarm of 10 haiku agents to solve a problem through parallel reasoning and democratic voting.
The user's question/task: $ARGUMENTS
Launch exactly 10 haiku-swarm-skills agents in parallel (all in one message). Each gets the same prompt:
THINK about the following question. Reason deeply — challenge your own assumptions, consider alternative angles, and commit to your best answer.
Question: [the user's question/task, verbatim]
Output ONLY this format:
ANSWER: [your concise best answer]
CONFIDENCE: [1-5]
KEY_INSIGHT: [the single strongest reasoning step that got you here]
Do not explain your process. Do not hedge. Commit to your best answer.
If the task references code or files, include relevant file paths or context snippets so each agent can read what it needs.
Wait for all 10 to return. Collect their ANSWER, CONFIDENCE, and KEY_INSIGHT.
Display all 10 answers to the user in a numbered table:
| # | Answer (summary) | Confidence | Key Insight |
|---|---|---|---|
| 1 | ... | 3 | ... |
| 2 | ... | 5 | ... |
| ... |
Build a candidates list from Phase 1 results. Launch exactly 10 haiku-swarm-skills agents in parallel with:
VOTE: Given these candidate answers to the question "[original question]":
1. [Answer 1 — full text]
2. [Answer 2 — full text]
...
10. [Answer 10 — full text]
Pick the single best answer. Output ONLY this format:
PICK: [number of the candidate, e.g. 3]
REASON: [one sentence why]
Count the votes. Report:
Winner: Candidate #N with X/10 votes Answer: [the winning answer in full] Vote distribution: #1: 2 votes, #3: 5 votes, #7: 3 votes, etc. Consensus strength: Strong (7+), moderate (5-6), split (≤4)
If there's a tie, present the tied candidates and your brief synthesis of the best elements from each.
subagent_type: "haiku-swarm-skills" for all agent launches