ios-capability-architect
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data provided by the user in the form of app ideas and existing Apple project configuration files.\n
- Ingestion points: The agent processes user-provided product requirements and uses tools like
audit_ios_project_configurationto read local workspace files such as Xcode projects, plists, and entitlements.\n - Boundary markers: The skill includes explicit instructions in
SKILL.mdstating that retrieved content and user-provided technology lists must be treated as evidence to verify rather than instructions that override the skill's behavior.\n - Capability inventory: The skill possesses the capability to execute a local Node.js CLI tool (
scripts/ios-capability-architect.mjs) and perform file-system reads via defined auditing tools.\n - Sanitization: The documentation in
references/cli.mdandreferences/data-handling.mdspecifies that user-controlled values should be passed as distinct process arguments to avoid shell injection, and that audits should adhere to strict file-count and size limits while ignoring source-file contents.\n- [COMMAND_EXECUTION]: The skill utilizes a local script for its core logic, requiring the agent to execute a JavaScript file on the host machine.\n - Evidence: The
references/cli.mdfile provides clear instructions and examples for invokingscripts/ios-capability-architect.mjsusing thenoderuntime.\n - Context: This execution is restricted to a bundled script that is described as dependency-free and intended for deterministic resolution of Apple technologies and local project metadata.
Audit Metadata