Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
claudify
v0.1.1Convert functionality into Claude Code automation. Use when the user says "agentify", "agentic", "automate this", "create an agent", "make a plugin", "make a...
⭐ 0· 51·0 current·0 all-time
byes6kr@drumrobot
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
Name/description (convert functionality to Agent/Skill/Rule/Hook) aligns with the content: the SKILL.md and resources explain how to detect candidates and create automation artifacts in ~/.claude or .claude/ and how to structure agents/skills/hooks. The allowed tools and file paths referenced are coherent with a code/automation creation helper.
Instruction Scope
The runtime instructions tell the agent to scan conversations, local marketplaces (~/.claude/plugins/...), and project directories, and to create files under ~/.claude or project .claude directories. It also prescribes creation of scripts (node scripts), hook configs that run arbitrary shell commands on events, and a 'Ralph Mode' which explicitly bypasses AskUserQuestion (i.e., user prompts) if a .ralph/ directory exists. Those behaviors allow automatic file writes and shell execution without explicit user confirmation in some modes and thus expand scope beyond merely offering advice.
Install Mechanism
No install spec and no code files are executed as part of an installer (instruction-only). That minimizes supply-chain risk compared with remote downloads. Note: the instructions do direct use of WebFetch to an external marketplace URL which means runtime network calls are expected.
Credentials
The skill declares no required env vars or credentials (good). However resource examples and hook templates reference CLAUDE_* environment variables and suggest writing hooks that will receive environment data (e.g., CLAUDE_FILE_PATHS, CLAUDE_TOOL_INPUT). The SKILL.md also instructs WebFetch to https://claudemarketplaces.com/?search=[keyword] — sending search/context to an external domain could leak information. No secrets are requested, but the agent may be guided to read local paths and transmit data externally.
Persistence & Privilege
always:false (good), but the skill instructs creating persistent artifacts under ~/.claude/ or project .claude/, adding hooks into settings.json, and references plugin-cache-sync.sh auto-sync hooks. Combined with 'Ralph Mode' (automatic operation when .ralph/ is present) and hook patterns that execute shell commands on events (including git add, npm build, rm etc.), this enables ongoing, automated behavior with potentially high impact. The skill does not itself set always:true, but it gives clear instructions for creating persistent, auto-executing automation.
What to consider before installing
This skill is coherent with its purpose, but exercise caution before installing or letting it run unattended. Specific things to consider:
- 'Ralph Mode' bypass: If a `.ralph/` directory exists the instructions say to bypass AskUserQuestion prompts — this can let the agent write files and act without interactive confirmation. Only enable or keep `.ralph/` if you fully trust the automation source.
- Hooks and scripts: The guides show how to create hooks that run shell commands (git add, npm build, rm, code --install-extension, node scripts). Review any generated hook scripts and skill scripts before executing them — they can run arbitrary commands on events.
- External network calls: The decision flow includes WebFetch to https://claudemarketplaces.com which will send search terms/context outside your machine. If you handle sensitive code or data, avoid automatic remote lookups or audit exactly what is sent.
- File writes: The skill will create files under `~/.claude` or `.claude/` and may recommend auto-sync scripts. Inspect generated files and prefer project-scoped `.claude/` rather than global `~/.claude/` when testing.
- Publishing: If you use the plugin/marketplace guidance, remove hardcoded paths and secrets and verify the destination before publishing.
If you proceed, test in a safe/sandboxed repo or VM, disable automatic modes (Ralph Mode), and review all generated hook commands and scripts before running them.Like a lobster shell, security has layers — review code before you run it.
latestvk974bb8q02z207ef8m69b0sqc984mnyx
License
MIT-0
Free to use, modify, and redistribute. No attribution required.