The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests various file types (Markdown, Text, JSON, etc.) from user-specified directories to seed the knowledge of a new subagent. It lacks boundary markers or sanitization logic to distinguish between data and instructions within these files.
Ingestion points: ~/.self-driving-agents/* and paths provided in the /hindsight-memory:create-agent command.
Boundary markers: Absent. The instructions do not warn the agent to ignore or escape instructions embedded in ingested content.
Capability inventory: The created agent has access to the full hindsight toolset including knowledge recall and injection, and the local file system via Read and Write (per the template instructions).
Sanitization: Absent.
[COMMAND_EXECUTION]: The skill uses Bash to list and read files within the ~/.self-driving-agents/ directory. This is used to gather seed content for agent creation.
[EXTERNAL_DOWNLOADS]: The documentation references the use of npx @vectorize-io/self-driving-agents install to prepare directories. This involves downloading and executing a Node.js package. As this package is provided by the skill's author (vectorize-io), it is documented here as part of the tool's intended lifecycle.

create-agent