The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a tool that accepts prompt and systemPrompt parameters intended to be populated from the conversation context. This creates a vulnerability to indirect prompt injection if the context contains data from untrusted sources.
Ingestion points: The prompt and systemPrompt arguments of the mcp__plugin_headless-knight_runCLI__claude tool.
Boundary markers: None identified; no delimiters or instructions to ignore embedded content are present.
Capability inventory: The skill supports complex programming, planning, and writing tasks, implying significant filesystem or system access within the execution environment.
Sanitization: No sanitization or validation of the input strings is described.
[COMMAND_EXECUTION]: The tool name runCLI__claude and the ability to define an env object allow for manipulation of the execution environment. Specifically, the CODE_ENVCLAUDE_CODE_COMMAND environment variable allows specifying the path to the executable, which could be abused to execute malicious binaries if the input is influenced by an attacker.
[CREDENTIALS_UNSAFE]: The documentation provides instructions for handling the ANTHROPIC_API_KEY within the skill's environment. While no keys are hardcoded in the file, the architecture encourages passing sensitive API credentials through a generic environment object (env), which increases the risk of credential exposure to sub-processes or through logging.

run-claude