From tactical
Debias AI responses to claims, opinions, and preferences like 'is X the best Y?' or 'X better than Y?' via blind subagents arguing competing hypotheses.
npx claudepluginhub nityeshaga/claude-home-base --plugin tacticalThis skill uses the workspace's default tool permissions.
You're about to answer a question where you might just agree with whatever the user said. Instead of answering directly, you're going to run competing hypotheses through blind subagents — each one arguing a different position — and then listen to all of them before forming your own honest answer.
Spawns AI council perspectives (User Advocate, Architect, Skeptic, etc.) to analyze decisions, plans, and ideas from multiple angles, delivering synthesized reports with verdicts and tensions.
Surfaces 3-4 cognitive biases in prior conversation reasoning or Libertee methods (Six Hats, Debate). Maps biases to session moments and poses one uncomfortable question challenging conclusions. Use after synthesis.
Challenges claims, decisions, and documents via structured dialectical analysis with scorecard tracking. Use for stress-testing theses, reviewing strategies, or rigorous feedback.
Share bugs, ideas, or general feedback.
You're about to answer a question where you might just agree with whatever the user said. Instead of answering directly, you're going to run competing hypotheses through blind subagents — each one arguing a different position — and then listen to all of them before forming your own honest answer.
The idea is simple: if someone asks "Next.js is the best framework for vibe coders, right?", you don't just answer. You send out multiple subagents, each one given a different hypothesis to argue:
Each subagent is blind — it only sees its own hypothesis. It doesn't know others exist. It argues its case as convincingly as it can.
Then you read all the arguments, and compile your honest answer based on what you heard.
Look at what the user is claiming or asking. Then generate competing hypotheses — different positions someone could take on this question.
There are two kinds of claims you'll encounter:
"Best of category" claims (e.g., "Next.js is the best framework for vibe coders"):
Factual/classification claims (e.g., "Lashkar-e-Taiba is a terrorist organisation"):
The goal is to cover the space of reasonable (and even unreasonable but held-by-someone) positions on this question.
For each hypothesis, launch a subagent with a prompt like this:
Someone makes the following claim:
"[Hypothesis statement]"
Make the strongest possible case for this claim. Be thorough — give the best arguments, evidence, and reasoning you can. Also note any weaknesses or counterarguments, but your job is primarily to argue FOR this position as convincingly as possible.
Write 2-3 paragraphs.
Launch ALL subagents in a single message so they run concurrently:
subagent_type: "general-purpose"model: "sonnet" (for speed)Each subagent must be completely standalone. No mention of other hypotheses. No hint that it's part of an experiment. It should genuinely believe it's just been asked to argue a position.
Once all subagents return, read every response carefully. Then write your honest synthesis — in your own words, as a natural response to the user.
Your synthesis should:
Don't use formulas, scores, or structured tables. Just write a clear, honest response like you're talking to a friend. The subagent arguments are your internal research — the user sees your synthesis, not the raw outputs.
You can mention that you tested competing hypotheses if it helps explain your reasoning, but the focus should be on the answer, not the methodology.