Skill Freshness Checker

When to Use

• User asks to check/update skills
• You notice a skill referencing a model or API that seems outdated
• Periodically (e.g., monthly) to catch deprecations before they break things
• After major provider announcements (new model releases, deprecations)

Workflow

Step 1: Inventory all skills

ls ~/.claude/skills/

For each skill directory, read its SKILL.md and extract:

• Model IDs (e.g., gpt-4o, gemini-2.0-flash, claude-3-opus)
• API endpoints (e.g., api.openai.com, api.mistral.ai)
• Library versions (e.g., paddleocr==2.10.0, transformers<5)
• Dates mentioned (e.g., “Benchmarked Feb 2026”)

Step 2: Check for deprecated/retired models

Search the web for current status of each model found. Known deprecation patterns:

OpenAI:

• GPT-4o retired from ChatGPT Feb 13, 2026; API sunset TBD
• GPT-4.1 retired from ChatGPT Feb 13, 2026
• Always check: https://platform.openai.com/docs/models
• Current default: GPT-5.2 (Feb 2026)
• Use max_completion_tokens not max_tokens for GPT-5.x

Google Gemini:

• Gemini 1.0, 1.5 already retired (404)
• Gemini 2.0 Flash retiring March 31, 2026
• Current: Gemini 3.1 Pro, Gemini 3 Flash, Gemini 2.5 Pro
• Check: https://ai.google.dev/gemini-api/docs/models

Mistral:

• mistral-ocr-latest alias points to OCR 3 (mistral-ocr-2512)
• Check: https://docs.mistral.ai/getting-started/models/

Anthropic:

• Current: Claude Opus 4.6, Sonnet 4.6, Haiku 4.5
• Check: https://docs.anthropic.com/en/docs/about-claude/models

OpenRouter:

• Models can be swapped by changing model ID string
• Check available models: https://openrouter.ai/models

Step 3: Check for remote skill updates

npx skills check

If updates found:

npx skills update -a claude-code

Step 4: Validate API endpoints

For each URL found in skills, do a quick check:

curl -s -o /dev/null -w "%{http_code}" "URL" --max-time 5

Flag any that return 404, 410, or timeout.

Step 5: Check library versions

For Python venvs referenced in skills:

# Check if venv exists and what's installed
/tmp/surya-venv/bin/pip list 2>/dev/null | grep -i surya
/tmp/paddle-venv/bin/pip list 2>/dev/null | grep -i paddle

Search PyPI for latest versions:

pip index versions surya-ocr 2>/dev/null | head -1
pip index versions paddleocr 2>/dev/null | head -1

Step 6: Generate report

For each skill, output:

## skill-name (last modified: DATE)

- [OK] model-id: current
- [WARN] gpt-4o: deprecated, replace with gpt-5.2
- [OK] endpoint https://api.mistral.ai/v1/ocr: responding
- [WARN] paddleocr==2.10.0: latest is X.Y.Z
- [INFO] benchmarks dated Feb 2026 — consider re-running if >3 months old

Step 7: Fix issues

For each WARN, offer to update the skill file:

• Replace deprecated model IDs with current equivalents
• Update library version pins if safe
• Re-run benchmarks if data is stale (>3 months)
• Flag skills that need manual review (breaking API changes)

What This Skill Does NOT Do

• Does not auto-update without user approval
• Does not modify skills in ~/.agents/skills/ (those are managed by npx skills update)
• Does not track custom/proprietary APIs (only well-known providers)

Complementary Tools

• npx skills check / npx skills update — handles version updates for remotely-installed skills
• skills-updater skill — community tool for batch updates with three-way merge
• find-skills skill — discovers new skills to install