The Agent Skills Directory

[SAFE]: The skill consists exclusively of markdown instructions, templates, and behavioral guidelines for AI agents. It does not include any scripts, executables, or automated persistence mechanisms.
[PROMPT_INJECTION]: The skill instructions create a surface for indirect prompt injection by directing the agent to process external URLs from user prompts and incorporate out-of-band context messages into the task flow.
Ingestion points: External URLs provided in prompts and system-level harness injections (SKILL.md, references/agent-behavior.md).
Boundary markers: No specific boundary markers or isolation protocols are mandated for external content ingestion.
Capability inventory: The agent is expected to possess shell and file system capabilities to perform the underlying tasks being logged (references/template.md).
Sanitization: There are no instructions for sanitizing or filtering content retrieved from external sources.
[COMMAND_EXECUTION]: The skill instructs the agent to execute simple shell commands such as cat to display the completed friction log to the user. This is a legitimate functional requirement for providing visible task output.
[SAFE]: All external URL references point to well-known and trusted service platforms (Vercel) for the purpose of log visualization.

friction-log