The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to install the CLI using npx -y @mvanhorn/printing-press-library, which involves downloading and executing code from the NPM registry at runtime.
[EXTERNAL_DOWNLOADS]: The skill utilizes external repositories for binary installation, specifically github.com/mvanhorn/printing-press-library via the go install command.
[DATA_EXFILTRATION]: The CLI tool supports a --deliver webhook:<url> flag, which creates a functional surface for sending command outputs, potentially containing job data or local metadata, to arbitrary external HTTP endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill implements a persistent 'learning loop' that presents a significant injection surface:
Ingestion points: Data enters the context via the Ashby API and the local learn_events store (containing playbooks and notes).
Boundary markers: The skill relies on natural language instructions for the agent to follow a decision tree and manually strip PII.
Capability inventory: The system can execute various shell commands through the ashby-pp-cli wrapper and write to local storage.
Sanitization: There is a 'PII rule' mentioned to manually strip personal identifiers before teaching the store, but no automated sanitization is described.
[DYNAMIC_EXECUTION]: The 'playbook' system allows the agent to execute sequences of commands (steps) defined in JSON files. These commands are dynamically assembled at runtime using slot substitution based on current and historical query data.

pp-ashby