The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions recommend installing the CLI using npx -y @mvanhorn/printing-press-library install learn-loop-example, which fetches content from the npm registry.
[REMOTE_CODE_EXECUTION]: The installation process involves executing code from the @mvanhorn/printing-press-library package using npx.
[COMMAND_EXECUTION]: The skill operates by executing shell commands for resource management. Additionally, the 'playbook' system instructs the agent to replay a sequence of stored CLI commands to fulfill user requests.
[PROMPT_INJECTION]: The 'Self-learning loop' exposes an indirect prompt injection surface (Category 8) by instructing the agent to treat stored, potentially user-influenced data as authoritative guidance.
Ingestion points: The agent retrieves data and instructions from a local SQLite database via the recall command (e.g., Playbook.steps and Notes).
Boundary markers: The instructions do not provide boundary markers or warnings to isolate or ignore malicious instructions that might be embedded in the recalled content.
Capability inventory: The agent has permissions to execute arbitrary bash commands (Read Bash tool) and can send data to external webhooks using the CLI's --deliver flag.
Sanitization: There is no mechanism described for sanitizing or validating the 'recalled' playbooks or notes before the agent is instructed to treat them as 'ground truth' and 'replay the steps.'

pp-learn-loop-example