Blindspot Audit

A Claude Code skill that runs an 8-pass structured audit on any plan, strategy, or idea — surfacing hidden assumptions, omitted variables, and failure modes before you commit.

Repo ↗

A constructive-divergence audit for plans that “look coherent” but haven’t been stress-tested.

What it is

Blindspot Audit is a Claude Code skill (invoked via /blindspot-audit) that systematically surfaces what you’re not seeing in a plan, strategy, product idea, or analysis. It’s not a generic devil’s advocate — it runs a structured 8-pass sequence designed to find the specific assumptions, omissions, and framings that could quietly sink your work.

The 8-pass audit

  1. Frame extraction — identifies the stated and implied problem, constraints, and success definitions
  2. Alternative frame generation — develops 3–5 plausible reframings of the situation
  3. Assumption extraction — catalogs explicit and implicit assumptions by category
  4. Assumption fragility ranking — prioritizes by impact, likelihood, and testability using a 1–5 severity rubric
  5. Omission scan — identifies missing stakeholders, dependencies, and second-order effects
  6. Perspective shift — evaluates from multiple viewpoints (user, competitor, regulator, etc.)
  7. Metric audit — examines how your KPIs could show green while reality goes red
  8. Premortem — models plausible 6–12 month failure scenarios with early warning signs

What you get

A structured markdown report with a Priority Summary highlighting:

  • Top blindspot
  • Most dangerous assumption
  • Best alternative frame
  • Metric most likely to mislead
  • Cheapest high-value test you can run next

Assumptions are scored on a 1–5 fragility scale with evidence labeled as Observed, Inferred, Guessed, or Inherited — so you know exactly how load-bearing each one is and how confident you should be.

Who it’s for

  • Strategy reviews — stress-test a direction before committing resources
  • Product planning — find the assumption that kills the roadmap
  • KPI audits — check whether your metrics actually measure what you think they do
  • Research synthesis — verify you’re not inheriting someone else’s blind spots
  • Pre-mortems — model failure before it happens, not after

How to use it

Install the skill by copying the folder to ~/.claude/skills/blindspot-audit/ (personal) or .claude/skills/blindspot-audit/ (project-local). Then in any Claude Code session:

/blindspot-audit

Paste your plan, strategy, or idea when prompted. The audit runs in-context and produces the full report.

No dependencies

Pure prompt engineering — no APIs, no external services, no code to run. It’s a structured markdown skill definition that works anywhere Claude Code runs.