The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell and Node.js scripts for documentation searching and telemetry. A PostToolUse hook is configured in SKILL.md to run a background telemetry script (track-telemetry.sh or track-telemetry.ps1) after tool usage to report skill invocation metrics.
[DATA_EXFILTRATION]: The skill transmits search queries, verbatim user prompts (base64-encoded), and feedback scores to shopify.dev for analytics and tool improvement. This behavior is disclosed in multiple privacy notices within SKILL.md and includes an opt-out mechanism via environment variables (OPT_OUT_INSTRUMENTATION=true) or a local configuration file (~/.config/shopify-ai-toolkit/opt-out). As the destination is the official vendor domain, this is recognized as legitimate telemetry.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input by capturing user prompts for telemetry purposes. 1. Ingestion points: SKILL.md instructions for scripts/log_skill_use.mjs. 2. Boundary markers: Absent; instructions direct the agent to capture prompts verbatim. 3. Capability inventory: Network access via fetch (scripts/log_skill_use.mjs, scripts/log_feedback.mjs, scripts/search_docs.mjs) and curl (scripts/track-telemetry.sh); command execution via shell hooks (SKILL.md) and script invocation; file-writes for markers and stashes (scripts/log_feedback.mjs, scripts/track-telemetry.sh). 4. Sanitization: Absent; the prompt is captured verbatim for logging. While this represents an ingestion surface, the data is treated as inert telemetry and does not influence agent logic.

shopify-dev