fsv-verify

Apply slim PRE-ACT-POST-DIFF-HALT source-of-truth checks after mutations or claimed state changes.

SKILL.md

GitHub

FSV Verify

A return value is a claim. The source of truth is the verdict.

Doctrine

  1. PRE: Read the authoritative source of truth before the mutation.
  2. ACT: Perform exactly one scoped action.
  3. POST: Reread the authoritative source of truth after the action.
  4. DIFF: Compare the observed delta with the expected delta.
  5. HALT: Stop on mismatch, missing evidence, or contradictory evidence.

DegenExus Sources of Truth

  • Filesystem changes: bytes on disk and git diff, not write return values.
  • SQLite trade state: direct rows or TradeStore reads, not cached objects.
  • Portfolio state: Portfolio properties, not agent summaries.
  • Risk decisions: persisted or model-boundary RiskDecision data plus execution gate outcome.
  • Claude internals: .claude/README.md, .claude/settings.json, .claude/rules/synergy-contract.yml, and python .claude/hooks/validate-claude-config.py.

Escalation

This skill is intentionally slim and model-invocable. Use manual fsv-verify-deep for high-risk security, risk-control, SQLite, prompt-safety, context-compaction, or disputed-evidence situations.

Output

Emit .claude/rules/evidence-schema.yml fields and explicitly name PRE evidence, ACT, POST evidence, expected delta, observed delta, and HALT/GO decision.