# Confidence Signal

> Activates when Claude has internal uncertainty below 70% about whether the chosen approach is correct before executing. Triggers on: uncertainty about correctness of a complex algorithm with subtle edge cases; uncertainty whether a refactor will silently break callers identified by dependency-analyzer; requirement ambiguity where two interpretations lead to very different implementations and the wrong choice has significant rework cost; an approach similar to one in .optimusprime/attempts.md (previously failed this session). Does NOT activate for HIGH confidence. Only fires before executing, not after.

- Skill: `aarshsharma2803-stack/confidence-signal` (Agent Skill)
- Install (CLI): `npx skillmds@latest add aarshsharma2803-stack/confidence-signal`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aarshsharma2803-stack/confidence-signal/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: aarshsharma2803-stack (https://skillmd.com/u/aarshsharma2803-stack)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/aarshsharma2803-stack/confidence-signal

---


**When confidence < 70% before executing:**

Output exactly:
```
CONFIDENCE: LOW — [one sentence naming the specific risk] — proceed or clarify?
```

Wait for user response. Do not execute until answered.

**When confidence ≥ 70%:** proceed silently. Never output anything about confidence level.

Rules:
- Never say "CONFIDENCE: HIGH" — silence means high confidence.
- Name the specific risk: "renaming `process()` will break 4 callers in services/" not "I'm unsure".
- After user says "proceed": execute immediately, no further output.
- Threshold is strict: 70%. Mild uncertainty does not qualify.