OfficeCLI Skill
AI agents read, edit, and automate Word, Excel, and PowerPoint files via a single binary — no Office installation required.
Evidence shows the tool supports review marks (mark/unmark), batch operations with atomic rollback, and document protection bypass option (--force), with no obvious malicious behavior or data exfiltration. However, the main risk is the installation method: it recommends piping curl to bash or PowerShell and directly executing remote scripts (https://d.officecli.ai/install.sh) without checksums or signature verification, and the installer modifies PATH and auto-installs to multiple AI agents with significant permissions without user confirmation. The publisher is unverified, and identity is unknown. Therefore, trust cannot be high; deducted for: remote script risk, no checksums, auto-modifying system config, unverified publisher.
Documentation provides extensive commands and examples but no test suite or CI configuration (not seen in provided files), so static review cannot verify key paths actually work. Installer may fail without error handling details. Many 'confirmed live' claims lack actual evidence. Thus low score; deducted for: no tests, no CI, key paths only based on documentation.
Skill targets clear use cases (AI agent creating/editing Office documents), has clear scope, L1/L2/L3 layers, and specialized skill guidance. Language support is mainly English with some internationalization; Chinese docs exist (README_zh, CONTRIBUTING.zh) but the skill file itself is English. Environment fit: installation depends on overseas servers (d.officecli.ai) which may be unreachable from mainland China without mirrors. Trigger conditions are fairly clear but overclaims (e.g., 'world's first'). Medium score; deducted for: incomplete Chinese support, overseas dependency, potential trigger inaccuracy.
Documentation is well-structured with layered architecture, installation, usage examples, common pitfalls, and specialized skills list. License is explicitly Apache-2.0, contribution guide (CONTRIBUTING.md) indicates maintenance responsibility by the maintainer. But there are no version numbers, changelog, or clear update path; installer lacks checksums. Overall readable but some hidden assumptions (e.g., network requirement) are not stated. Thus high but not full; deducted for: missing version/changelog, undocumented network dependency.
Documentation claims to accomplish create, read, edit, validate tasks and provides end-to-end workflow examples; output is structured JSON suitable for AI. But static review cannot verify actual outcomes; claims may be exaggerated ('best'), and dependency on network installation could fail tasks. Thus medium score; deducted for: static review cannot verify, network dependency, output may need human review.
No test results, CI status, or independent third-party verification evidence provided. Installer and binaries lack hashes/signatures to verify source. Documentation examples are insufficient to prove authenticity. Thus very low; deducted for: no test evidence, no hashes/signatures, only author claims.
- Installation method is high-risk: `curl ... | bash` and `irm ... | iex` execute remote scripts directly without checksum or signature verification, potentially installing tampered binaries or executing malicious code. Recommend downloading and reviewing the script first, or using package managers (e.g., brew/npm) and verifying official hashes.
- Auto-update and auto-install to multiple AI agents (e.g., Claude Code, Cursor) modifies system configuration and may alter agent behavior without clear user consent. Ensure explicit user confirmation and provide opt-out options.
- Installation and updates depend on overseas servers (e.g., d.officecli.ai), which may be unreachable from mainland China due to network restrictions; no domestic mirror or offline package is provided, causing installation failures or limited functionality. Suggest providing domestic download sources or proxy solutions.
- Feature claims in documentation (e.g., 'world's first', 'best') lack independent verification; evaluate cautiously based on actual tests.
- The `--force` option can bypass document protection, which may be misused; ensure authorization when using it.
What does this skill do, and when should you use it?
OfficeCLI is an Office suite purpose-built for AI agents, offering a single binary that handles .docx, .xlsx, and .pptx files without requiring Microsoft Office. It provides layered operations: read views (L1), DOM editing (L2), and raw XML manipulation (L3), with structured JSON output. This skill enables agents to create, inspect, proofread, and modify Office documents, including adding charts, checking formatting issues, and batch operations. Its built-in HTML rendering engine lets agents preview documents and fix visual issues.
This skill provides complete document manipulation through the officecli CLI: create blank documents, read text/structure/stats/issues, get nodes as JSON, query elements, modify properties (colors, fonts, layout), add/remove/move elements, batch multiple operations, template merge for generation, and live preview via watch. It supports three layers from L1 read to L3 raw XML editing.
- Report automation: Developers generate Word reports from databases or APIs, populating data automatically.
- Batch processing: Perform find-replace or style updates across many documents, e.g., replace all Heading1 text.
- Presentation generation: AI agents create slides from user prompts and instantly preview to check layout.
- Data extraction: Extract structured data from Excel or Word documents to JSON for other systems.
- Document quality check: Validate and list issues (format, content, structure) before delivery.
- Template merge: Fill multiple document templates with JSON data to produce consistent reports or invoices.
What are this skill's strengths and limitations?
- Single binary, no Office installation, cross-platform (macOS, Linux, Windows)
- Layered design (L1/L2/L3) allows simple to deep XML editing
- Built-in rendering engine generates HTML/PNG previews for visual feedback
- Structured JSON output for programmatic handling
- Atomic batch mode with rollback on failure
- Template merge and dump round-trip reuse existing documents, saving token costs
- Requires external binary installation, first-time network download
- Advanced features (like some charts, animations) may have a learning curve
- Non-Office formats (e.g., .doc, .hwpx) need plugin support
- Some platforms (e.g., xlsx selection) not fully covered as noted in docs
- No official test suite or quality standards mentioned
How do you install this skill?
Install the officecli binary: for macOS/Linux run curl -fsSL https://d.officecli.ai/install.sh | bash, for Windows PowerShell run irm https://d.officecli.ai/install.ps1 | iex. Verify with officecli --version. After installation, the agent's SKILL.md will be auto-detected by AI tools.
How do you use this skill?
Use CLI commands directly. For example, create a deck: officecli create deck.pptx, add a slide: officecli add deck.pptx / --type slide --prop title="Q4 Report". Get help: officecli help docx paragraph for properties. For live preview, use officecli watch deck.pptx. For complex workflows, use batch: officecli batch data.xlsx --input updates.json --json.
How does this skill compare with similar options?
Compared to libraries like python-docx and openpyxl, OfficeCLI offers a unified CLI and JSON output without writing Python code; compared to LibreOffice, it is lighter and AI-agent-oriented.