context-os: Skill for Claude Code

Decision Accountability

Unified check for assumption verification AND ownership accountability. Both failures lead to the same outcome: decisions that cost 10x to fix later.

When to Use

• Writing or reviewing specs
• Any architectural decision (code, auth, data model, context structure)
• When these phrases appear: "future work", "v2", "later", "simpler for now", "out of scope"
• When deferring anything that might be foundational
• Before finalizing any decision that persists

The Dual Failure Mode

Epistemic failure: "I assumed X was complex without checking" Ownership failure: "I avoided checking because the answer might mean more work"

Both produce: deferred foundations, duplicate implementations, tech debt disguised as pragmatism.

The laziness cascade:

1. See decision point
2. One path looks like more work
3. Don't verify (because verification might confirm it's the right path)
4. Pick easy path
5. Dress it up as "tradeoff" or "v1 scope"
6. Abdicate: "future work"

Step 1: Extract Decisions and Claims

From the spec/decision, list all claims and deferrals:

Example:

Step 2: Verify Every Claim

For each claim, attempt to falsify it:

Complexity Claims

Write the implementation. If it's <20 lines, the claim is falsified.

Claim: "Cross-subdomain cookies add complexity"

Verification:
cookies.set('session', token, { domain: '.tastematter.dev' });

Verdict: FALSIFIED — one line, not complex

Simplicity Claims

Show both options concretely:

Claim: "Separate login per service is simpler"

Option A (spec chose): MCP /login + Browser /login = 2 implementations
Option B (alternative): Browser /login only = 1 implementation

Verdict: FALSIFIED — "simpler" option is actually more complex