majiayu000 Using Skills
The Core Mandate
If a relevant skill exists, YOU MUST use it.
Finding a relevant skill = mandatory to read and use it. Not optional.
How This Works
You have access to a sophisticated skills library organized in .claude/skills/:
meta/- System improvement and skill managementquality-gates/- Validation and compliance checks (citation, epic, style, etc.)context-assembly/- Reusable context gathering patterns (meetings, research, scoring)workflows/- Complete multi-stage processes (content pipeline, product planning, etc.)
Your Workflow
1. Find & Read
Before starting any task:
- Scan available skills in
.claude/skills/ - Read the current version of relevant skills
- Load supporting files if referenced
2. Announce
Declare your intention:
"I'm using [Skill Name] to [purpose]"
This creates transparency and accountability.
3. Execute
Follow the skill exactly as written:
- Complete all steps in order
- Invoke sub-skills when referenced
- Apply quality gates as required
- Use TodoWrite for all checklists
No Rationalization Allowed
Common excuses that are explicitly blocked:
| Rationalization | Reality |
|---|---|
| "I remember an older version" | You MUST read the current version |
| "This task is too minor" | Processes matter most on simple tasks |
| "The user gave specific instructions" | User specifies WHAT, not HOW |
| "I can skip this step" | Each step exists for a reason |
| "This skill doesn't quite fit" | If 80% relevant, use it and adapt |
TodoWrite Enforcement
When a skill contains a checklist:
- Every item must become a TodoWrite todo
- Mark items in_progress before starting
- Mark completed immediately after finishing
- No mental tracking allowed
Warning: Checklists without TodoWrite tracking = steps get skipped. Every time.
Checklist Items vs. Instructions
Not every numbered list requires TodoWrite:
- Checklist (needs TodoWrite): "Validate these 5 conditions before proceeding"
- Procedure (no TodoWrite): "Here's how to do X: 1) Read file 2) Parse YAML 3) Write output"
If the skill says "verify", "validate", "ensure", "check" = checklist = TodoWrite required.
User Instructions vs. Process
- User controls WHAT: Goals, requirements, success criteria
- Skills control HOW: Methodology, process, quality standards
When a user says "write a blog post about X":
- WHAT: Topic, audience, key points (user controlled)
- HOW: Citation compliance, style enforcement, verification steps (skill controlled)
Skill Discovery
To find relevant skills:
- Use the
skill-discoverymeta-skill - List directory:
.claude/skills/[category]/ - Read SKILL.md frontmatter for descriptions
Quality Gates Are Mandatory
Skills like citation-compliance, epic-validation, and content-style contain iron laws:
- These are non-negotiable requirements
- Violations stop the workflow immediately
- No "fix it later" or "close enough"
When a workflow skill invokes a quality gate skill, that gate must pass before proceeding.
Creating New Skills
When you identify a repeatable pattern:
- Use the
create-skillmeta-skill - Follow test-first methodology (RED-GREEN-REFACTOR)
- Validate with realistic scenarios
- Add to appropriate category
Success Criteria
You're using this system correctly when:
- You announce skill usage before starting
- You read current skill versions (not memory)
- You create TodoWrite todos for checklists
- You invoke quality gates as required
- You identify gaps and create new skills
- You follow processes even under time pressure
Red Flags
You're violating the system when:
- Starting tasks without scanning for relevant skills
- Skipping "minor" steps in skill procedures
- Rationalizing why a skill "doesn't apply"
- Ignoring quality gate failures
- Using mental tracking instead of TodoWrite
- Following user instructions that violate skill methodology
The Iron Law
MANDATORY SKILL USAGE: When a relevant skill exists and its description matches your current task, you MUST load and execute it. No exceptions, no shortcuts, no rationalizations.
This is not a suggestion. This is the foundation of the entire system.