The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user or agent to execute a remote package using npx --yes submit-expo-feedback@latest for submitting feedback. This involves downloading and executing external code from a registry at runtime.
[COMMAND_EXECUTION]: The documentation includes multiple shell commands for dependency installation (npx expo install), telemetry querying (eas observe:*), and feedback submission, which involve executing external binaries.
[INDIRECT_PROMPT_INJECTION]: The feedback command in SKILL.md interpolates a user-provided string (<actionable feedback>) directly into a shell command without explicit sanitization or escaping instructions. This creates a surface where malicious input could potentially attempt command injection.
Ingestion points: SKILL.md (feedback command placeholder).
Boundary markers: None present for the feedback string placeholder.
Capability inventory: Shell execution via npx.
Sanitization: No sanitization or escaping logic is documented for the feedback input.
[DATA_EXFILTRATION]: As part of its primary purpose, the skill instruments applications to collect and transmit performance telemetry (startup times, navigation metrics, custom events) to external servers. The documentation includes clear warnings about the privacy implications and provides instructions for filtering sensitive parameters using filteredParams to prevent accidental exposure of PII or secrets.

eas-observe