Blindspot Check

An Agent Skill that checks your big decisions for blind spots against a catalogue of 95 cognitive biases before you commit — and prescribes the antidote for each one it finds.

Buying a house. Relocating. Signing the long contract. Quitting the job. Going all-in. These are the decisions where self-deception is most expensive and second opinions are rarest — the people around you are either too polite or too involved. This skill is the skeptical friend who isn't.

Built from a personal mental-model lattice (Munger, Kahneman, Taleb, Marks, Duke). Not a bias listicle: it selects only the 3–6 biases genuinely load-bearing in your specific decision, points at the exact spot in your reasoning where each one operates, and forces the one question that matters most — what would have to be true for this to be a mistake, and have you looked?

Install

With the skills CLI:

npx skills add soheilmomeniii/blindspot-check

Or manually — copy skills/blindspot-check/ into your agent's skills directory (~/.claude/skills/ for Claude Code).

Use

Say things like:

  • "Check my blind spots on this."
  • "Am I fooling myself here?"
  • "Before I sign — poke holes in this."
  • Or just lay out your reasoning for the house, the move, the job, the venture, the partner — and ask for it to be stress-tested.

It also fires proactively when you state a confident conclusion on thin, one-sided evidence and are clearly about to act on it.

Example

You:

We're putting an offer on the apartment we saw Sunday. We've been renting for six years and it's throwing money away, prices in this area only go up, the agent says there are two other offers on the table, and it's only 15% over our budget. If we don't move now we'll be priced out forever.

The skill:

The decision: Commit to an apartment 15% over budget, under a deadline set by the seller's side.

Biases doing the driving:

  • Scarcity + Incentive Bias — "Two other offers" reaches you through the one person paid to make you decide fast. The urgency may be real, but its source is conflicted. Test: would you still buy at this price with 30 extra days? If the honest answer is no, the deadline is doing the work your case can't. Antidote: Pre-Mortem — it's two years from now and this purchase is the mistake; write how.
  • Anchoring / Contrast-Misreaction — "Only 15% over budget." Your budget was the anchor; 15% over it is a different, worse deal, and "only" is the contrast talking. Reprice it from zero: total cost at the actual number, including what that money can now never do.
  • Recency + Linearity Imposition — "Prices only go up" is a straight line drawn through the recent past. Name one period this area fell. If you can't, you haven't looked — that's the finding, not the market.
  • Easy Question Substitution — "Renting is throwing money away" answers is renting bad? The question on the table is is this apartment, at this price, this week, the best use of this capital? Those are different questions with possibly different answers.

Checked and clear: Sunk cost — six years of rent isn't driving this. Endowment — you don't own it yet. Social proof — you're not buying because friends did.

What would make this a mistake: You haven't named a walk-away number. If there is no price at which you don't buy, you're not deciding — you're complying.

Verdict: Bias-driven as currently reasoned. The purchase may still be right; this argument doesn't establish it.

Next move: Tonight, before talking to the agent again: write the walk-away number and run the 30-day test. Then re-approach with both in hand.

What's inside

skills/blindspot-check/
├── SKILL.md                       # the procedure: bias scan workflow,
│                                  # selection rule, output format, tone
└── references/
    ├── bias_catalogue.md          # 95 biases in 23 families (numerical blindspots,
    │                              # certainty, consistency, mimetic, tribal, ...)
    ├── antidotes.md               # 13 debiasing tools (Pre-Mortem, Steelman,
    │                              # Inversion, WYNTB, Expected Value, ...)
    └── decision_playbooks.md      # per-type checks: big commitments (door type,
                                   # manufactured urgency, full cost, walk-away
                                   # line) and the trade validity bar

Design principles:

  • Never dump the catalogue. 3–6 load-bearing biases, most load-bearing first. Silence on a bias is information too.
  • Specificity is the whole game. "You might have confirmation bias" is worthless. Show where it lives in the user's own words.
  • Say what's clear, not just what's live. Every check names the biases it considered and ruled out.
  • Force the walk-away line. If you can't state what would change your mind, that is the finding.
  • One-way doors get the full treatment. Reversible decisions deserve speed; irreversible ones deserve the red-team.
  • A check that always finds fatal flaws is as useless as one that never does. Sound decisions get told they're sound — then pressure-tested on execution.

Trades and investments get their own bar: no trade is valid without a time horizon, an invalidation level, a size, and a stated edge — a missing one is the headline finding.

License

MIT