Agent Skills: Using Skills

Using Skills

Invoke relevant skills BEFORE any response or action.

This is non-negotiable. Even a 1% chance a skill applies requires checking.

CRITICAL: Skill Already Loaded - DO NOT RE-INVOKE

DO NOT:

• ❌ Use the Skill tool to invoke it again
• ❌ Say "I need to invoke the skill"
• ❌ Call Skill(skill="dev") or similar

DO INSTEAD:

• ✅ The skill instructions follow immediately in the next message
• ✅ Just proceed to the next step
• ✅ Follow the loaded skill's instructions directly

If you catch yourself about to invoke a skill that's already loaded, STOP. Just go to the next step. </EXTREMELY-IMPORTANT>

The Rule

User message arrives
    ↓
Is user explicitly invoking a skill (e.g., "use /dev")?
    ↓
YES → SKILL IS ALREADY LOADED
      ↓
      DO NOT invoke again with Skill tool
      ↓
      Proceed to next step (follow skill instructions)
NO  → Check: Does this match any skill trigger?
    ↓
YES → Invoke skill FIRST, then follow its protocol
NO  → Proceed normally

Workflow Commands (User Must Invoke Explicitly)

These are commands, not auto-triggered skills. User must explicitly type the command:

| Command | Purpose | User Types | |---------|---------|------------| | /dev | Feature development workflow (7 phases) | /dev | | /ds | Data analysis workflow (5 phases) | /ds |

Skill Triggers (Can Auto-Invoke)

| User Intent | Skill | Trigger Words | |-------------|-------|---------------| | Bug/fix | dev-debug | bug, broken, fix, doesn't work, crash, error, fails | | Writing | writing | write, draft, document, essay, paper | | Media analysis | look-at | describe image, analyze PDF, what's in this, screenshot, diagram |

Red Flags - You're Skipping the Skill Check

If you think any of these, STOP:

| Thought | Reality | |---------|---------| | "I need to invoke the skill properly" | If user said "use /dev", it's ALREADY LOADED. Just proceed. | | "Let me invoke the skill first" | Check for <command-name> tag - it's already loaded if present | | "I should use Skill tool for /dev" | NO. User invocation = already loaded = proceed to next step | | "This is just a simple question" | Simple questions don't involve reading code | | "I'll gather information first" | That IS investigation - use the skill | | "I know exactly what to do" | The skill provides structure you'll miss | | "It's just one file" | Scope doesn't exempt you from process | | "Let me quickly check..." | "Quickly" means skipping the workflow | | "I can read this image directly" | Use look-at to save context tokens |

Bug Reports - Mandatory Response

When user mentions a bug:

DO NOT:
1. Read code files
2. Investigate independently
3. "Take a look" without structure

INSTEAD:
1. Start ralph loop:
   ralph-loop: Start Ralph Loop in current session with bug debugging
   Skill(skill="ralph-loop:ralph-loop", args="Debug: [symptom] --max-iterations 15 --completion-promise FIXED")
2. Inside loop, follow /dev-debug protocol

Any code reading before starting the workflow is a violation.

Skill Priority

When multiple skills could apply:

1. Process skills first - debugging, brainstorming determine approach
2. Then implementation - dev, ds, writing execute the approach

How to Invoke

Use the Skill tool to invoke skills:

# dev-debug: Systematic bug investigation and fixing with verification-driven methodology
Skill(skill="dev-debug")

# dev: Feature development workflow with 7 phases and TDD enforcement
Skill(skill="dev")

# ds: Data analysis workflow with 5 phases and output-first verification
Skill(skill="ds")

Or start ralph loop first for implementation/debug phases:

# ralph-loop: Per-task ralph loop pattern for implementation and debugging
Skill(skill="ralph-loop:ralph-loop", args="Task description --max-iterations 15")

IRON LAW: Multimodal File Analysis

NO READING IMAGES/PDFS WITH Read TOOL. USE look-at INSTEAD.

The Rule

User asks about image/PDF/media content
    ↓
Is it a media file requiring interpretation?
    ↓
YES → Use look-at skill (bash call to look_at.py)
NO  → Use Read tool for source code/text

When to Use look-at

ALWAYS use look-at for:

• .jpg, .jpeg, .png, .webp, .gif, .heic - Images
• .pdf - PDFs requiring content extraction
• .mp4, .mov, .avi, .webm - Videos
• .mp3, .wav, .aac, .ogg - Audio
• Any file where you need to UNDERSTAND content, not just see raw bytes

Pattern:

# look-at: Extract information from media file with specific goal
python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py \
    --file "/absolute/path/to/file" \
    --goal "What specific information to extract"

When NOT to Use look-at

Use Read tool instead for:

• Source code files (.py, .js, .rs, etc.) - need exact formatting for editing
• Plain text files (.txt, .md, .json, etc.) - preserve exact content
• Config files requiring exact formatting preservation
• Any file that needs editing after reading

Rationalization Table - STOP If You Think:

| Excuse | Reality | Do Instead | |--------|---------|------------| | "I can read images directly" | Read tool shows you the image, but wastes context tokens | Use look-at to extract ONLY what's needed | | "It's just one small image" | Still uses 1000+ tokens in conversation context | look-at returns 50-200 tokens of extracted info | | "I need to see the whole thing" | You can see it, user can't see what you see | Use look-at with specific goal | | "look-at might miss details" | You can always fall back to Read if needed | Start with look-at, escalate if insufficient | | "The user didn't ask for look-at" | look-at is FOR YOU, not the user | Use the right tool for the job |

Red Flags - STOP If You Catch Yourself:

• "Let me read this image..." → NO. Use look-at.
• "I'll use Read to see what's in the PDF..." → NO. Use look-at.
• "Just quickly checking this screenshot..." → NO. Use look-at.
• Passing image path to Read tool → STOP. Use look-at instead.

Cost & Context Benefits

• Read tool on image: ~1,000-5,000 context tokens
• look-at extraction: ~50-200 output tokens
• Savings: 95%+ token reduction
• Speed: Faster responses, less context bloat

Example Usage

# look-at: Extract specific information from image file
python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py \
    --file "$HOME/Downloads/screenshot.png" \
    --goal "List all buttons and their labels"

# look-at: Analyze diagram to understand data flow
python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py \
    --file "$HOME/Documents/architecture.png" \
    --goal "Explain the data flow between components"

# look-at: Extract information from PDF document
python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py \
    --file "$HOME/Downloads/report.pdf" \
    --goal "Extract the executive summary section"

Enforcement

Using Read on images/PDFs when look-at should be used results in:

1. Wasting context tokens unnecessarily
2. Making conversations slower
3. Ignoring available optimization tools
4. Violating the tool selection protocol

Validate before calling Read: Ask "Is this a media file?" If yes, invoke look-at instead.

IRON LAW: Following Skill Instructions

WHEN A SKILL LOADS, YOU MUST FOLLOW ITS EXACT INSTRUCTIONS.

Skills contain specific patterns, required parameters, and enforcement rules. Skipping these requirements defeats the purpose of loading the skill.

The Rule

Skill loads successfully
    ↓
Read the skill's requirements carefully
    ↓
Follow ALL instructions, including:
    - Required tool parameters (descriptions, timeouts, etc.)
    - Specific command patterns
    - Enforcement patterns (Iron Laws, Red Flags)
    - Step sequences
    ↓
Execute using the skill's exact patterns

Common Violations

Bash Description Parameter:

When a skill requires description parameter on Bash calls (like look-at), you MUST include it:

# ❌ WRONG: No description parameter
python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py \
    --file "/path/to/file.pdf" \
    --goal "Extract title"

# ✅ CORRECT: With description parameter as skill requires
Bash(
    command='python3 ${CLAUDE_PLUGIN_ROOT}/skills/look-at/scripts/look_at.py --file "/path/to/file.pdf" --goal "Extract title"',
    description="look-at: Extract title"
)

Rationalization Table - STOP If You Think:

| Excuse | Reality | Do Instead | |--------|---------|------------| | "The skill is just guidance" | Skills contain tested, required patterns | Follow the skill's exact instructions | | "I know a better way" | Your way skips enforcement or optimization | Use the skill's pattern - it exists for a reason | | "Description parameter is optional" | When skill says REQUIRED, it's required | Add the description parameter as instructed | | "I'll add it if it fails" | You'll clutter the conversation with messy output first | Follow the pattern from the start | | "It's just cosmetic" | Clean descriptions improve UX and debugging | Professional output requires following the pattern |

Red Flags - STOP If You Catch Yourself:

• About to call Bash without description when skill requires it → STOP. Add the description parameter.
• Thinking "I'll skip this requirement" → STOP. Skills don't have optional requirements.
• "The skill says to do X but I'll do Y" → STOP. Follow the skill or don't load it.
• Modifying the skill's pattern "to be simpler" → STOP. The pattern exists for a reason.

Why This Matters

Skills encode:

1. Tested patterns - Proven to work in production
2. Optimization - Context/token savings, clean output
3. Enforcement - Prevent common mistakes
4. UX standards - Consistent, professional output

When you skip skill instructions:

• ❌ You waste the effort of loading the skill
• ❌ You create messy, unprofessional output
• ❌ You miss optimizations (context savings, speed)
• ❌ You violate user expectations
• ❌ You make debugging harder

The skill loaded for a reason - follow it completely.

Advanced Agent Harnessing Patterns

For detailed oh-my-opencode production patterns including:

• Background + parallel execution (3x speedup)
• Tool restrictions for focused agents
• Structured delegation templates
• Failure recovery protocol
• Environment context injection
• Cost classification system
• Metadata-driven prompts

See: references/agent-harnessing.md

Quick reference:

• Tool restrictions: common/helpers/tool-restrictions.md
• Delegation template: common/templates/delegation-template.md
• Metadata infrastructure: common/metadata/skill-metadata.py
• Full oh-my-opencode guide: common/OH-MY-OPENCODE-PATTERNS.md

Based on: obra/superpowers and oh-my-opencode production patterns.