The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a remote script from a GitHub repository and executes it by piping the content directly to the shell. Evidence: curl -sL "https://raw.githubusercontent.com/flobby41/claude-code-ultimate-guide/main/examples/scripts/audit-scan.sh" | bash -s -- --json.
[COMMAND_EXECUTION]: The skill runs multiple shell commands to inspect the local environment and configurations. Evidence: Uses cat, ls, jq, and which to query system state and configuration settings.
[DATA_EXFILTRATION]: The skill reads sensitive configuration files that contain permission settings and server configurations. Evidence: Reads ~/.claude/settings.json and ~/.claude.json. These files can contain sensitive tool permissions and custom service configurations.
[INDIRECT_PROMPT_INJECTION]: The skill fetches a remote markdown guide which it uses as its primary knowledge base, creating a surface where instructions embedded in the remote file could influence the agent's behavior. 1. Ingestion points: ultimate-guide.md fetched via curl in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: File system access and shell command execution. 4. Sanitization: No sanitization of the remote content before processing.

diagnose