Delegate

Delegate

Help the user decide: should this task be fully automated, AI-assisted, or kept human? Not everything should be automated. This skill provides a structured decision framework.

When to Activate

Manual triggers:

• "Should I automate this?"
• "AI or human?"
• "Can Claude do this, or should I?"
• "Delegate this"
• "Is this worth automating?"

Auto-detect triggers:

• User hesitates about whether to automate something
• User describes a task with mixed automation potential (partly automatable,

partly needs judgment)

• User asks "can Claude handle this?" about a task with ambiguous complexity

The Evaluation Framework

For each task, evaluate across five dimensions. Score each 1-5.

1. Repetitiveness (1-5)

How often does this task recur with a similar pattern?

• 5 = Daily or more, nearly identical each time
• 4 = Weekly, same structure with minor variations
• 3 = Monthly, moderate variations
• 2 = Quarterly or irregular, significant variations
• 1 = One-off or rarely recurring

2. Judgment Required (1-5, inverted — high score = low judgment)

How much human judgment, creativity, or nuance is needed?

• 5 = Zero judgment — purely mechanical (data formatting, file conversion)
• 4 = Minimal judgment — clear rules cover 90%+ of cases
• 3 = Moderate judgment — some edge cases need human review
• 2 = Significant judgment — context-dependent decisions
• 1 = Deep judgment — requires empathy, politics, creativity, or taste

3. Error Tolerance (1-5)

What happens if the automation gets it wrong?

• 5 = Errors are trivially fixable, no consequences
• 4 = Errors cause minor inconvenience, easily caught
• 3 = Errors are noticeable, require some cleanup
• 2 = Errors cause real problems — wrong data sent, incorrect decisions
• 1 = Errors are catastrophic — legal, financial, safety, or reputational damage

4. Data Sensitivity (1-5, inverted — high score = low sensitivity)

How sensitive is the data involved?

• 5 = Public or non-sensitive data
• 4 = Internal but non-confidential
• 3 = Business-confidential
• 2 = Personally identifiable information (PII) or financial data
• 1 = Regulated data (HIPAA, SOX, classified) or credentials

5. Current Time Cost (1-5)

How much time does this task consume?

• 5 = Multiple hours per occurrence
• 4 = 1-2 hours per occurrence
• 3 = 30-60 minutes per occurrence
• 2 = 15-30 minutes per occurrence
• 1 = Under 15 minutes per occurrence

The Decision Matrix

Calculate the Delegation Score = average of all five dimensions.

| Score Range | Recommendation | What It Means | |---|---|---| | 4.0-5.0 | AUTOMATE FULLY | Build it and forget it. No human in the loop. | | 3.0-3.9 | AI-ASSIST | Claude does the heavy lifting, human reviews before finalizing. | | 2.0-2.9 | HUMAN + AI TOOLS | Human drives, Claude is a tool (research, drafting, analysis). | | 1.0-1.9 | KEEP HUMAN | This task needs human judgment, empathy, or accountability. Don't automate. |

Output Format

Present the evaluation as:

# Delegation Assessment: [Task Name]

## Scores
| Dimension | Score | Reasoning |
|---|---|---|
| Repetitiveness | X/5 | [One sentence why] |
| Judgment Required | X/5 | [One sentence why] |
| Error Tolerance | X/5 | [One sentence why] |
| Data Sensitivity | X/5 | [One sentence why] |
| Current Time Cost | X/5 | [One sentence why] |

## Delegation Score: X.X/5

## Recommendation: [AUTOMATE FULLY / AI-ASSIST / HUMAN + AI TOOLS / KEEP HUMAN]

### Why
[2-3 sentences explaining the recommendation, highlighting the dimension(s)
that most influenced the decision]

### If Automating
[Specific recommendation for how to automate, what tools to use]

### If Keeping Human
[What the human should focus on, and where AI can still help at the margins]

### Risk to Watch
[The single biggest risk of following this recommendation, and how to mitigate it]

Rules

1. Never recommend full automation for tasks scoring below 2 on Error Tolerance

or Data Sensitivity. Some tasks are too risky regardless of other scores.

1. Be honest when automation isn't the answer. Saying "keep this human" is a

valid, valuable recommendation. Don't force automation where it doesn't fit.

1. When recommending AI-ASSIST, be specific about the split. "Claude drafts,

you review" is better than "use AI to help."

1. Consider the user's technical level. A non-technical user can't maintain a

custom Python script. Recommend tools they can actually use.

1. If the scores are borderline (e.g., 2.8-3.2), present both options and let

the user decide. Acknowledge the ambiguity.

Chaining

• automate-audit → delegate: When the audit surfaces a task, use delegate

to decide if it should actually be automated

• delegate → decompose: If the recommendation is AI-ASSIST, decompose the

task to find which sub-steps are automatable

• delegate → prompt-architect: If the recommendation is AUTOMATE, build the

prompt to do it

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ryankolean
• Source: ryankolean/summit-claude-skills
• License: MIT

Install and usage instructions live in the source repository linked above.