The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the claude CLI with parameters derived from user input and conversation history. \n- Evidence: SKILL.md contains the execution pattern: claude --bg --name \"<descriptive name>\" \"<handoff summary>\". \n- Risk: There is a potential for shell command injection if the user-provided <descriptive name> or the generated <handoff summary> contain shell metacharacters (such as backticks or semicolons) that are not correctly escaped by the model during string interpolation. \n- [PROMPT_INJECTION]: The skill ingests untrusted data to generate a summary that seeds the prompt of a new agent session. \n- Ingestion points: Conversation history and user-supplied arguments (SKILL.md). \n- Boundary markers: The skill lacks explicit instructions for the model to use delimiters or sanitization techniques to prevent the generated summary from containing instructions that could override the next agent's behavior. \n- Capability inventory: The skill triggers the creation of a new agent session with shell execution capabilities. \n- Sanitization: The instructions include a mandate to redact sensitive credentials, but do not address the risk of embedded malicious instructions within the processed data.

claude-handoff