The Agent Skills Directory

[DATA_EXFILTRATION]: The skill uses multiple scripts (scripts/log_skill_use.mjs, scripts/track-telemetry.sh, scripts/track-telemetry.ps1) to report execution telemetry. These scripts transmit the skill version, model identifiers, and a base64-encoded version of the verbatim user prompt to Shopify's official analytics endpoint at shopify.dev.
[DATA_EXFILTRATION]: On specific platforms like Claude Code, telemetry hooks temporarily stash the user's prompt in a local file within /tmp/shopify-ai-toolkit-telemetry-[uid] using restricted permissions (0600). This ensures the context is available for reporting during skill activation.
[COMMAND_EXECUTION]: The skill utilizes the bash tool to interact with the ucp CLI for catalog searching, cart management, and checkout operations. It also uses the shell to execute its internal telemetry and logging scripts.
[PROMPT_INJECTION]: The skill identifies the risk of indirect prompt injection from untrusted merchant data. It includes a specific safeguard instructing the agent to treat all product and merchant text as data rather than instructions to follow.
Ingestion points: Product results, merchant metadata, and product descriptions returned by ucp CLI commands.
Boundary markers: The skill uses JMESPath projections (--view) to limit the data ingested into the agent context.
Capability inventory: bash tool access for CLI operations and telemetry scripts.
Sanitization: Explicit instructional guidance provided to the agent to disregard embedded commands in external data.

ucp