← ClaudeAtlas

prompt-confidence-gatelisted

Evaluate every user prompt on every turn against a minimum confidence threshold before executing actions. Read the threshold from AGENTS.md or GEMINI.md, ask the user if unset, and use the Python scorer to evaluate candidate intents when a prompt is ambiguous.
rachmanzz/skills · ★ 0 · AI & Automation · score 75
Install: claude install-skill rachmanzz/skills
# Prompt Confidence Gate Use this skill on **every turn** before executing any action, tool call, or code change. Its purpose is to verify that the user's prompt is sufficiently unambiguous and meets the required minimum confidence threshold prior to execution. Read the supporting references as needed: - Read [evidence-and-assumption-control.md](references/evidence-and-assumption-control.md) before making claims about user intent, thresholds, or confidence scores. - Read [confidence-rubric.md](references/confidence-rubric.md) for threshold configuration syntax, scoring dimensions, and CLI usage of `scripts/score_prompt.py`. ## Per-turn execution workflow Follow these four steps on every turn without exception: ### 1. Resolve the minimum confidence threshold 1. Inspect `AGENTS.md` and `GEMINI.md` in the active workspace (and user-level agent configuration if present) for an explicit `confidence_threshold` directive (for example, `confidence_threshold: 0.85` or `confidence_threshold: 85%`). 2. If a valid threshold is defined in `AGENTS.md` or `GEMINI.md`, use it as the authoritative minimum threshold for the current turn. 3. If no threshold is defined in `AGENTS.md` or `GEMINI.md` and none has been provided by the user for this session, **halt execution** and ask the user to specify the minimum confidence value (`0.00`–`1.00` or `0%`–`100%`) required before execution. ### 2. Identify candidate intents when ambiguity exists 1. Inspect the user's prompt for ambiguity sig