The Agent Skills Directory

[SAFE]: The skill consists solely of natural language instructions for the AI agent to follow. It does not invoke any tools, perform network operations, or interact with the local file system. The logic explicitly instructs the agent to output the result to the chat for the user to copy, rather than saving to a file, which maintains human-in-the-loop control.
[NO_CODE]: The skill does not include any scripts, executables, or external dependencies. It relies entirely on the underlying model's reasoning capabilities to process the provided instructions.
[PROMPT_INJECTION]: The skill is a summarization tool that ingests conversation history, which constitutes untrusted data. This presents a potential surface for indirect prompt injection, where malicious instructions hidden in a previous conversation could be included in the 'handoff' brief. However, the risk is minimal as the output is intended for manual review and copying by the user.
Ingestion points: Conversation context processed during the creation of the handoff brief.
Boundary markers: The output is structured using a specific Markdown template defined in the instructions.
Capability inventory: The skill has no capabilities to execute commands, write files, or make network requests.
Sanitization: There are no explicit instructions to sanitize the conversation history for potentially malicious content before including it in the output.

handoff