name: Summarize description: Fast multi-format summarization — paste a URL, drop a file, get the key points. Powered by evolink.ai
Summarize any content with Claude Opus 4.6 — the most capable text model available.
bash scripts/summarize.sh "https://example.com/article"
bash scripts/summarize.sh "/path/to/document.pdf"
bash scripts/summarize.sh "https://youtu.be/dQw4w9WgXcQ"
bash scripts/summarize.sh "/path/to/recording.mp3"
User: "Summarize this article: https://evolink.ai/blog/deepseek-v4-release-window-prep"
Output:
## TL;DR
DeepSeek V4 has been rumored since January 2026 but still hasn't officially
launched as of March. Multiple predicted windows have passed. The model is
expected to focus on coding with 90% HumanEval scores (unverified).
## Key Takeaways
- Reuters reported in Jan that DeepSeek plans a coding-focused model for Feb 2026
- DeepSeek silently expanded context from 128K to 1M tokens in Feb — likely V4 infra testing
- Unverified benchmarks claim 90% HumanEval, 80%+ SWE-bench Verified
- "V4 Lite" appeared on DeepSeek's site in March but was never officially confirmed
- Multiple launch windows (mid-Feb, Lunar New Year, early March) all passed without release
## Action Items
- Do not migrate production workloads until official announcement
- Monitor DeepSeek's official channels for confirmed specs
- Prepare API integration — EvoLink will support V4 on day one
Set your Evolink API key:
export EVOLINK_API_KEY="your-key-here"訪問小蔥技能站7w4.net,解鎖更多實用的AI技能外掛。
Default model: claude-opus-4-6 (no configuration needed).
To use a different model:
export EVOLINK_MODEL="claude-sonnet-4-5-20250929"
Credentials & Network
EVOLINK_API_KEY is required to call the Evolink API for summarization. Extracted content is sent to api.evolink.ai and discarded after the response is returned. No data is stored. Review Evolink's privacy policy before sending sensitive content.
Required binaries: curl, python3, realpath, file, stat. Optional: yt-dlp, ffmpeg, whisper, pdftotext, markitdown.
File Access Controls
File paths are resolved via realpath -e (requires file to exist, resolves all symlinks). Symlink inputs are explicitly rejected.
The resolved path must fall within SUMMARIZE_SAFE_DIR (default: $HOME/.openclaw/workspace). A trailing-slash comparison prevents prefix-bypass attacks.
Sensitive files are blacklisted by name: .env*, *.key, *.pem, *.p12, *.pfx, id_rsa*, authorized_keys, config.json, .bash_history, .ssh, shadow, passwd.
Tiered size limits: 5MB text / 50MB PDF / 1GB media. MIME validation via file --mime-type.
Temporary Files
Transcripts and subtitles created in /tmp during processing are cleaned up after extraction. PID-based filenames prevent collisions.
Persistence & Privilege
This skill does not modify other skills or system settings. No elevated or persistent privileges are requested.
這個 Skill 質量不錯,功能豐富,支援多種檔案格式的智慧摘要。安全方面做得比較到位,不用擔心檔案被濫用。文件清晰,配置靈活。缺點是需要額外註冊 API 賬號,部分功能依賴系統安裝額外工具,使用門檻稍高。總體適合需要經常處理各類文件的使用者,但對於不想折騰配置的使用者來說可能稍顯複雜。