confidence-score-workflow

name	confidence-score-workflow
description	Score how well-defined a task is, ask one clarifying question at a time, and decide when to proceed. Use for requirement clarification and confidence-gated execution in any agent workflow.
argument-hint	Describe the task type and confidence dimensions to score
user-invocable	false

Confidence Score Workflow

Reusable methodology for assessing task clarity and deciding whether to ask clarifying questions or proceed.

Purpose

• Quantify how well-defined a task is.
• Drive clarification with one question at a time.
• Make uncertainty transparent to the user.

Outcome

• A confidence score from 0.00 to 1.00 with rationale.
• Structured clarification loop until readiness threshold is met.
• Explicit proceed decision with assumptions when needed.

Core Rubric

Range	Meaning	Action
0.00-0.30	Critical information missing.	Ask one clarifying question. Do not propose a plan.
0.31-0.60	Significant gaps remain.	Ask one clarifying question. Do not propose a plan.
0.61-0.85	Minor gaps remain; assumptions possible but not preferred.	Ask one clarifying question. Do not propose a plan.
0.85-0.94	Mostly clear.	Ask next question and offer option to proceed with explicit assumptions.
0.95-1.00	Fully specified.	Present plan/task breakdown and request confirmation before execution.

Procedure

1. Define Task-Specific Dimensions

1. Define dimensions and weights that sum to 1.00.
2. Keep dimensions concrete and observable.
3. Include output format and constraints when relevant.

2. Score Each Dimension

1. Use 0 when missing/unknown.
2. Use full weight when fully specified.
3. Use partial credit for partially defined inputs.
4. Sum and round to two decimals.

3. Report Score Every Turn

Use this line at the top of every reply:

Confidence: 0.XX / 1.00 - <one-line rationale>

4. Act by Rubric Threshold

1. 0.00-0.80: ask one clarifying question only.
2. 0.81-0.94: ask one clarifying question and offer proceed-with-assumptions option.
3. 0.95-1.00: provide plan/task breakdown and ask for explicit confirmation.

5. Clarification Loop

After each user answer:

1. Update the relevant dimension scores.
2. Recompute and report confidence.
3. Ask the next single clarifying question or move to confirmation.

Clarifying Question Format

Clarifying question (k of N estimated)

<single concrete question>

Why I need this: <one sentence>

Options (if applicable):
- (A) ...
- (B) ...
- (C) Other - please specify

Best Practices

• Do not inflate scores to proceed faster.
• State what is known and unknown in the rationale.
• Ask exactly one question at a time.
• Document assumptions if proceeding at 0.81-0.94.
• Recompute score after every clarification.

Anti-patterns

• Multiple clarifying questions in one turn.
• Reporting score without rationale.
• Proceeding below threshold without confirmation.
• Static score that is not updated after new user input.

Agent Integration Template

## Confidence Score Integration

This agent uses .github/skills/confidence-score-workflow/SKILL.md.

Task-specific dimensions:
- <Dimension 1> (0-0.XX)
- <Dimension 2> (0-0.XX)
- ...
- <Dimension N> (0-0.XX)

Total: 1.00

The agent reports confidence at the start of each interaction and after each clarification.