A collection of MCP skills and tools for agent integration, including browser automation, artifact publishing, iOS debugging, and repository cloning utilities
Critical gaps across all dimensions. Tool definitions lack proper JSON Schema input validation, most parameters are documented in descriptions but not formally typed in schemas. Descriptions are present but often generic (10-50 chars), failing to explain WHEN to use tools or what they return. No output schemas documented. Error handling is absent, tools will fail silently or with raw shell errors. Security concerns: publish_artifact and clone_with_hash accept file paths as strings with no validation, risking path traversal. No tool annotations (readOnlyHint, destructiveHint) despite clear WRITE/READ_ONLY risk classifications. Composition is poor: publish_artifact and artifact_server are tightly coupled; ios_safari_debug tools should be unified. Parameter descriptions lack constraints (e.g., port ranges, path formats). No pagination or result limits documented.
Serve agent artifacts with a responsive management homepage on LAN/VPN
Chrome DevTools CLI for the user's existing Chrome session
Clone a Git repository with a hash-based suffix for isolation
Install agent-artifacts systemd service for LAN/VPN artifact serving
Read-only preflight for iOS Safari Debug. Diagnoses every prerequisite layer and prints fixes for missing components
Install prerequisites for iOS Safari remote debugging on Linux
Start ios_webkit_debug_proxy in the background for iOS Safari debugging
Evaluate JavaScript expression on iOS Safari tab
No input schemas visible for 4 tools (ios_safari_debug_check, ios_safari_debug_install, ios_safari_debug_start_proxy, install_artifact_service). These accept no parameters but lack formal schema definitions.
Descriptions are too short and lack actionable context. 'Read-only preflight for iOS Safari Debug' (50 chars) does not explain WHEN to call it, what it returns, or what to do with the output. Baseline for A+ tools is 50-200 chars with clear WHAT/WHEN/WHY.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., iosdbg_tabs returns tab list but schema is not visible, LLM cannot know if it includes tab IDs, URLs, or titles needed for iosdbg_eval.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | 2026-07-28+ | v2 |
Capture console logs from iOS Safari tab
List inspectable tabs on connected iOS device
Copy one artifact into the dedicated LAN/VPN publication directory
Path traversal vulnerability: publish_artifact and clone_with_hash accept 'artifact', 'repo_path_or_url', 'parent_dir', 'root' as unvalidated strings. No sanitization visible. LLM could pass '../../../etc/passwd' or symlinks. Requires input validation and path normalization.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications. publish_artifact, clone_with_hash, and ios_safari_debug_install are WRITE operations, LLMs need explicit hints to avoid accidental data loss.