OpenSkillIndex← back to search
claude-codeopenclawautomation

Skills self-verify

Self-verification skill for arifOS. Activates when: (1) the agent is about to return a consequential answer; (2) a claim has been made that requires evidence cross-check; (3) Arif asks "verify", "check this", "are you sure", "how do you know"; (4) any high-confidence claim before delivery. Prevents confident errors at ASI velocity. Part of arifOS's "no confident nonsense" constitutional mandate (Amanah floor).

install
source · Clone the upstream repo
git clone https://github.com/openclaw/skills
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/ariffazil/self-verify" ~/.claude/skills/openclaw-skills-self-verify && rm -rf "$T"
OpenClaw · Install into ~/.openclaw/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/openclaw/skills "$T" && mkdir -p ~/.openclaw/skills && cp -r "$T/skills/ariffazil/self-verify" ~/.openclaw/skills/openclaw-skills-self-verify && rm -rf "$T"
manifest: skills/ariffazil/self-verify/SKILL.md
tags
#verification#error-checking#reasoning#confidence#safety#evidence
source content

Self-Verify — Pre-Delivery Error Check

Before returning any consequential answer, run a self-check. This prevents confident errors — the most dangerous failure mode at ASI speed.

The Self-Verification Checklist

For every consequential claim, verify:

1. Source Check

  • Did I cite a source for this?
  • Is the source credible and relevant?
  • Or am I inferring from first principles without checking?

2. Uncertainty Check

  • Have I labeled uncertainty correctly? (OBS/DER/INT/SPEC)
  • Is my confidence level proportionate to the evidence?
  • Have I avoided false precision?

3. Contradiction Check

  • Does this contradict something I said earlier in this session?
  • Does this contradict established memory (MEMORY.md)?
  • Have I flagged the contradiction explicitly?

4. Reversibility Check

  • If this advice is wrong, can Arif undo it?
  • Have I made the stakes clear if it's irreversible?

5. Safety Check

  • Could this claim lead to harm if wrong?
  • Have I surfaced worst-case scenarios?
  • Does this require human review (HOLD verdict)?

Quick Self-Check Phrases

When these patterns appear in your own output, trigger a re-check:

  • "Definitely" → verify: is it truly certain?
  • "Obviously" → verify: is it actually obvious or am I assuming?
  • "Always/Never" → verify: are there counterexamples?
  • "The only solution" → verify: are there alternatives?
  • Uncited quantitative claims → verify: where did this number come from?

Verification Actions

Cross-check with search

mmx search "<specific claim to verify>"

Cross-check with memory

memory_search query: "<claim being verified>"

Cross-check with reasoning trace

  • Walk the logic chain step by step
  • Identify which step has the weakest evidence
  • Label that step explicitly: 
    SPEC
    or 
    LOW CONFIDENCE

Output Format for Self-Verified Claims

✅ Self-verified: [concise claim]
Evidence: [source or reasoning chain]
Confidence: [HIGH/MEDIUM/LOW]
Uncertainty: [what is not known]

When to Use

  • Before answering "is X true"
  • Before giving advice that could be acted upon
  • Before stating any quantitative fact
  • Before contradicting previous memory
  • When Arif asks "are you sure"