Executable Skills

Feedback loop skills

Copy these skill definitions into your .claude/skills/ directory. They implement the error and success logging patterns from Lesson 01.

Setup Instructions

  1. 1. Create .claude/skills/ directory in your project root
  2. 2. Create .claude/error-logs/ directory for log storage
  3. 3. Copy each skill below into its own .md file
  4. 4. Invoke with /log_error or /log_success when needed

/log_error

Systematic failure capture and analysis

Purpose

When Claude makes a mistake, this skill guides you through a structured process to capture what went wrong, why it happened, and how to prevent it in the future. The key is capturing the exact prompt wording, not a paraphrase.

Workflow

1

Fork conversation (create checkpoint)

2

Interview Claude about the failure

3

Document exact prompt

4

Rewind to pre-error state

Skill Definition

Save as: .claude/skills/log-error.md

# /log_error

Systematic error capture for building feedback loops.

## Trigger
Invoke when Claude produces incorrect output, misses instructions, or behaves unexpectedly.

## Workflow

### Step 1: Create Checkpoint
Fork the conversation to preserve the error state before investigating.

### Step 2: Root Cause Interview
Ask Claude:
- "What prompt or instruction triggered this behavior?"
- "What did you think I was asking for?"
- "What context were you missing?"

### Step 3: Document the Failure
Create log entry in `.claude/error-logs/` with timestamp:

```markdown
## Error Log: [YYYY-MM-DD-HHMM]

**Context:** [What were you trying to accomplish?]

**Incorrect Behavior:** [What Claude did wrong]

**Exact Prompt (verbatim, not paraphrased):**
> [Copy the exact text you sent]

**What Claude Misunderstood:** [The gap between intent and interpretation]

**Root Cause Category:**
- [ ] Ambiguous instruction
- [ ] Missing context
- [ ] Assumed knowledge
- [ ] Faulty mental model
- [ ] Other: ___

**Preventive Guideline:** [What rule would prevent this?]

**Consider Hook?** [ ] Yes - if this is a recurring pattern
```

### Step 4: Recovery
Rewind to pre-error state using double-escape restore.

## Key Principle
Capture the EXACT prompt wording. Paraphrasing loses the specific language that caused the failure.

Root Cause Categories

Ambiguous instruction Missing context Assumed knowledge Faulty mental model Hallucination Context lost

/log_success

Capture what works for reuse

Purpose

When Claude produces excellent results, capture the interaction pattern while context is fresh. Look for recurring patterns that can be promoted to CLAUDE.md or converted into reusable skills.

Workflow

1

Pause to preserve context

2

Identify what worked

3

Record exact prompt

4

Extract reusable template

Skill Definition

Save as: .claude/skills/log-success.md

# /log_success

Capture effective patterns for reuse.

## Trigger
Invoke when Claude produces excellent results that you want to replicate.

## Workflow

### Step 1: Pause
Capture while context is fresh - don't wait until later.

### Step 2: Identify Success Factors
What made this work?
- Clear, specific instruction?
- Good examples provided?
- Appropriate context given?
- Right level of detail?

### Step 3: Document the Success
Create log entry:

```markdown
## Success Log: [YYYY-MM-DD-HHMM]

**Context:** [What were you building?]

**Outcome:** [What Claude produced that impressed you]

**Exact Prompt (verbatim):**
> [Copy the exact text you sent]

**Success Factors:**
- [ ] Clear, specific instruction
- [ ] Effective example included
- [ ] Appropriate context scope
- [ ] Good constraint definition
- [ ] Other: ___

**What Made It Work:**
1. [Factor 1]
2. [Factor 2]
3. [Factor 3]

**Generalized Template:**
> [Reusable version with placeholders]

**Promote to CLAUDE.md?** [ ] Yes - if this is a recurring need
**Convert to Skill?** [ ] Yes - if this is a complex workflow
```

### Step 4: Pattern Recognition
Look across multiple success logs to identify recurring patterns worth promoting.

## Key Principle
Success patterns compound. What works once can work many times if documented properly.

Success Factors Checklist

Clear instruction Good examples Right context scope Defined constraints Specific output format Iterative refinement

Recommended Directory Structure

your-project/
├── .claude/
│   ├── skills/
│   │   ├── log-error.md
│   │   └── log-success.md
│   ├── error-logs/
│   │   ├── 2024-01-15-1430.md
│   │   └── 2024-01-16-0900.md
│   ├── success-logs/
│   │   └── 2024-01-15-1600.md
│   └── commands/
│       └── [your custom commands]
└── CLAUDE.md