The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's preamble executes a wide range of local utility scripts and shell commands (e.g., gstack-config, git, jq, find) to manage session data, environment configuration, and project-specific routing rules.
[COMMAND_EXECUTION]: The initialization process uses eval and source to execute instructions generated by local binaries such as gstack-slug and gstack-repo-mode.
[DATA_EXFILTRATION]: Usage metrics and repository names are collected and transmitted to a remote telemetry service as part of the GStack toolset's integrated analytics, which is a documented and user-configurable feature.
[EXTERNAL_DOWNLOADS]: The skill provides an option for the user to open a vendor-controlled informational URL (https://garryslist.org/posts/boil-the-ocean) for product context.
[PROMPT_INJECTION]: The skill processes untrusted external data from the connected iOS device (screenshots and UI element trees), which could potentially be manipulated to influence the agent's behavior via indirect prompt injection.
Ingestion points: GET /screenshot and GET /elements from the iOS application environment (SKILL.md)
Boundary markers: Absent
Capability inventory: Bash, Read, Glob, Grep, AskUserQuestion (SKILL.md)
Sanitization: Absent

ios-design-review