The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the user to download and install a plugin from an unverified GitHub repository (Yeachan-Heo/oh-my-claudecode) and a global NPM package (oh-my-claude-sisyphus) whose name does not match the skill or the stated author.
[REMOTE_CODE_EXECUTION]: The installation of external plugins and NPM packages allows for the execution of arbitrary code within the agent's environment, which is compounded by the skill's high-privilege access to Bash, Write, and Edit tools.
[COMMAND_EXECUTION]: The skill performs several high-risk system operations, including the modification of the agent platform's experimental configuration (~/.claude/settings.json) and the execution of a background 'daemon' process (omc wait --start) for auto-resuming sessions.
[DATA_EXFILTRATION]: The skill's configuration instructions explicitly prompt users to provide sensitive API credentials, such as Telegram bot tokens and Discord webhooks, which are then stored and used by the tool.
[PROMPT_INJECTION]: As an orchestration layer, the skill ingests natural language instructions and distributes them to multiple agents with system access, creating an indirect prompt injection surface.
Ingestion points: Natural language task inputs via keywords like autopilot:, team:, and ralph:.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the orchestrator logic.
Capability inventory: Full access to Bash, Write, Edit, Read, Grep, and Glob across 32 specialized agents.
Sanitization: No validation or sanitization of user-supplied tasks before they are passed to the agent executor pipeline.

omc