The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (flight fares, route info, WiFi speeds) from various third-party providers.
Ingestion points: Data returned from search and status commands including flights, dates, explore, soar, award, and wifi (SKILL.md).
Boundary markers: The instructions do not define explicit delimiters or 'ignore embedded instructions' warnings for processing the results from these tools.
Capability inventory: The skill has the capability to execute shell commands, write to a local SQLite store (teach), and send data to external URLs (webhook) via the companion CLI.
Sanitization: There is no documentation of sanitization or escaping applied to external data before it is journaled or used to synthesize playbook candidates.
[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing a CLI binary and MCP server from remote repositories.
Evidence: Instructions suggest npx -y @mvanhorn/printing-press-library install and go install github.com/mvanhorn/printing-press-library/... (SKILL.md). These resources originate from the vendor's official repositories.
[DATA_EXFILTRATION]: The CLI binary includes a delivery sink feature that allows command output to be routed to external network endpoints.
Evidence: The --deliver webhook:<url> flag performs an HTTP POST of the command result to the specified URL (SKILL.md).
[DYNAMIC_EXECUTION]: The skill implements a 'playbook' system that dynamically reconstructs and executes command sequences.
Evidence: Playbooks are stored as JSON containing a steps array with cmd strings that are subject to slot substitution and execution in subsequent sessions (SKILL.md).

pp-flight-goat