The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use standard Unix commands (mkdir, mktemp) to prepare the file system for saving the handoff document.
Evidence: The instructions specify creating a directory with mkdir -p ./docs/handoffs/ and generating a unique filename using mktemp ./docs/handoffs/handoff-XXXXXX.md.
Context: These operations are restricted to the local project directory and are necessary for the skill's primary function of data persistence.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it summarizes untrusted conversation data into a document intended to guide the behavior of a subsequent agent.
Ingestion points: The skill reads the entire current conversation history to populate sections like 'Goal', 'Intent', and 'Stance'.
Boundary markers: The document uses Markdown headers (e.g., '## Next agent: start here') to separate sections, but does not explicitly warn the next agent to ignore embedded instructions within the summarized text.
Capability inventory: The skill has the capability to write files to the local file system.
Sanitization: No explicit sanitization or escaping of the conversation content is performed before it is interpolated into the handoff template.
Context: This risk is inherent to the skill's purpose of summarizing interactions, and the template includes a verification step where the current user reviews the inferred 'Intent' and 'Stance' before the file is saved.

create