From qe-framework
Creates, modifies, diagnoses, and benchmarks QE Framework skills. Handles skill creation from scratch, editing existing skills, and fixing skills that block or mishandle requests.
How this skill is triggered — by the user, by Claude, or both
Slash command
/qe-framework:Mcreate-skillThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Creates new skills and diagnoses/fixes existing ones.
Creates new skills and diagnoses/fixes existing ones.
reference/ subdirectory files.skills/<Name>/SKILL.md~/.claude/plugins/cache/inho-team-qe-framework/qe-framework/<ver>/skills/<Name>/SKILL.mdQE_CONVENTIONS.md. This file is the skill/agent registry that Claude reads at every session.| Component | Source | Cache |
|---|---|---|
| Skill | skills/<Name>/SKILL.md | ~/.claude/plugins/cache/.../skills/<Name>/SKILL.md |
| Agent | agents/<Name>/AGENT.md | ~/.claude/plugins/cache/.../agents/<Name>/AGENT.md |
| Hook | hooks/<name> | Same path (runs from source) |
| Global | ~/.claude/skills/<name>/SKILL.md | N/A |
| Local | .claude/skills/<name>/SKILL.md | N/A |
Simple (single file):
skills/Qname/
└── SKILL.md
Complex (with references — use when detailed content exceeds ~50 lines per topic):
skills/Qname/
├── SKILL.md ← Workflow, validation gates, quick reference
└── reference/ ← Detailed domain guides
├── topic-a.md
└── topic-b.md
Frontmatter:
---
name: Qskill-name
description: "Action. Trigger conditions. Distinction from similar skills."
metadata:
author: name
version: "1.0.0"
---
Choose a workflow pattern:
| Pattern | Use When | Structure |
|---|---|---|
| Sequential Steps | Procedural tasks | Step 0 → 1 → 2 → Validate → Confirm |
| Collaborative Phases | Documents, plans, reports | Structure agreement → Section-by-section → Visual polish → Export |
| Gate-and-Execute | Validation/review tasks | Read → Apply rules → Report |
Add validation gates if the skill produces quality-sensitive output:
## Validation (Required — Every Output)
1. Check A → condition = **FAIL**
2. Check B → condition = **FAIL**
Add component-based execution for large outputs:
Do not build everything at once. One unit at a time:
1. Decompose → 2. Prioritize → 3. Implement one → 4. User confirms → next
Make tool integrations explicit and mandatory, not optional suggestions.
End with Quick Reference + Never Use sections.
Scope: QE Framework internals only. Not for user project tasks.
| Situation | Example |
|---|---|
| Skill incorrectly blocks a valid request | Gate fires on a legitimate prompt |
| Skill routes to the wrong handler | Qcommit triggers on general commit questions |
| Trigger condition is too narrow or broad | Skill never fires, or fires on unrelated prompts |
| Decision rule uses wrong threshold | Word count limit rejects valid prompts |
| Workflow step produces incorrect behavior | Auto-approves when it should verify |
Collect:
Read the SKILL.md/AGENT.md. For each candidate rule:
| Class | Description | Fix Direction |
|---|---|---|
| Threshold | Numeric cutoff miscalibrated | Adjust with justification |
| Missing signal | Valid signal not recognized | Add signal type (not keyword) |
| Brittle encoding | Matches literals, not semantics | Rewrite to check intent |
| Scope mismatch | Guards the wrong thing | Redefine what rule checks |
| Missing case | Valid scenario unhandled | Add with reasoning |
If a fix requires adding more keywords to a list, the rule itself is likely wrong. Fix the principle, not the list.
Present before/after diff with:
Apply fix to both source and cache files (see File Locations above).
## Diagnosis
**Skill/Agent:** <name>
**Defect class:** <class>
**Root cause:** <one sentence>
## Fix
**Before:** <exact current text>
**After:** <proposed replacement>
**Why:** <reasoning>
## Test Cases
| Prompt | Expected | Reason |
|--------|----------|--------|
| ... | Pass | ... |
| ... | Gate | ... |
npx claudepluginhub inho-team/qe-framework --plugin qe-frameworkCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.