The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to construct and execute a shell command (claude --bg ...) using dynamically generated content from the conversation and user arguments. There are no instructions provided to the agent regarding escaping shell metacharacters (e.g., quotes, backticks, semicolons), which could lead to command injection if the summary contains malicious sequences.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by taking untrusted data (current conversation history and user-supplied arguments) and using it as the primary prompt for a new background agent.
Ingestion points: Conversation history and user-provided arguments in SKILL.md.
Boundary markers: None specified in the command template; the summary is passed as a raw string argument.
Capability inventory: Execution of the claude CLI tool to spawn new processes.
Sanitization: The skill explicitly instructs redacting sensitive data like API keys, which is a safety-positive measure, but it lacks instructions for handling embedded instructions or shell-breaking characters.
[DYNAMIC_EXECUTION]: The skill uses a template to generate executable command strings at runtime based on conversation context.

claude-handoff