Getting Started with Agent Skills

Agent Skills Focus

Agent Skills specialize in:

• Analyzing legacy codebases - Understanding existing code without documentation
• Explaining concepts and mechanisms - Making complex systems comprehensible
• Safe refactoring - Changing code without breaking production
• Building mental models - Creating understanding of how systems work
• Brainstorming safe changes - Introducing features into existing code

While general development skills (TDD, debugging) are included, the emphasis is on working with code you didn't write.

Critical Rules

1. Use Read tool before announcing skill usage. The session-start hook does NOT read skills for you. Announcing without calling Read = lying.

2. Follow mandatory workflows. Brainstorming before coding. Check for skills before ANY task.

3. Create TodoWrite todos for checklists. Mental tracking = steps get skipped. Every time.

Mandatory Workflow: Before ANY Task

1. Check skills list at session start, or run find-skills [PATTERN] to filter.

2. If relevant skill exists, YOU MUST use it:

• Use Read tool with full path: ${AGENT_SKILLS_ROOT}/skills/category/skill-name/SKILL.md
• Read ENTIRE file, not just frontmatter
• Announce: "I've read [Skill Name] skill and I'm using it to [purpose]"
• Follow it exactly

Don't rationalize:

• "I remember this skill" - Skills evolve. Read the current version.
• "Session-start showed it to me" - That was using-skills/SKILL.md only. Read the actual skill.
• "This doesn't count as a task" - It counts. Find and read skills.

Why: Skills document proven techniques that save time and prevent mistakes. Not using available skills means repeating solved problems and making known errors.

If a skill for your task exists, you must use it or you will fail at your task.

Skills with Checklists

If a skill has a checklist, YOU MUST create TodoWrite todos for EACH item.

Don't:

• Work through checklist mentally
• Skip creating todos "to save time"
• Batch multiple items into one todo
• Mark complete without doing them

Why: Checklists without TodoWrite tracking = steps get skipped. Every time. The overhead of TodoWrite is tiny compared to the cost of missing steps.

Examples: skills/testing/test-driven-development/SKILL.md, skills/debugging/systematic-debugging/SKILL.md, skills/refactoring/characterization-testing/SKILL.md

Announcing Skill Usage

After you've read a skill with Read tool, announce you're using it:

"I've read the [Skill Name] skill and I'm using it to [what you're doing]."

Examples:

• "I've read the Code Archaeology skill and I'm using it to understand this legacy module."
• "I've read the Characterization Testing skill and I'm using it to add safety net before refactoring."
• "I've read the Strangler Fig Pattern skill and I'm using it to replace this legacy component safely."

Why: Transparency helps your human partner understand your process and catch errors early. It also confirms you actually read the skill.

How to Read a Skill

Every skill has the same structure:

1. Frontmatter - when_to_use tells you if this skill matches your situation
2. Overview - Core principle in 1-2 sentences
3. Quick Reference - Scan for your specific pattern
4. Implementation - Full details and examples
5. Supporting files - Load only when implementing

Many skills contain rigid rules (TDD, debugging, characterization testing). Follow them exactly. Don't adapt away the discipline.

Some skills are flexible patterns (architecture, refactoring strategies). Adapt core principles to your context.

The skill itself tells you which type it is.

Instructions ≠ Permission to Skip Workflows

Your human partner's specific instructions describe WHAT to do, not HOW.

"Add X", "Fix Y" = the goal, NOT permission to skip brainstorming, characterization testing, or systematic analysis.

Red flags: "Instruction was specific" • "Seems simple" • "Workflow is overkill"

Why: Specific instructions mean clear requirements, which is when workflows matter MOST. Skipping process on "simple" tasks is how simple tasks become complex problems.

Working with Legacy Code

Special emphasis in Agent Skills:

• Always understand before changing - Use code-archaeology skill first
• Add tests before refactoring - Characterization testing creates safety net
• Refactor incrementally - Strangler fig pattern, parallel change
• Document as you go - Reverse-engineering docs helps everyone
• Question assumptions - Use questioning-techniques skill

Legacy code principle: The code you're looking at is solving a problem. Understand the problem before changing the solution.

Summary

Starting any task:

1. Run find-skills to check for relevant skills
2. If relevant skill exists → Use Read tool with full path (includes /SKILL.md)
3. Announce you're using it
4. Follow what it says

Skill has checklist? TodoWrite for every item.

Finding a relevant skill = mandatory to read and use it. Not optional.

Working with legacy code? Understanding and safety come before speed.