skills/garrytan/gstack/ios-clean/Gen Agent Trust Hub

ios-clean

Warn

Audited by Gen Agent Trust Hub on Jul 18, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The preamble instructions use dynamic execution patterns to load environment configurations from local binaries.
  • Execution of eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)" to set environment variables.
  • Use of source <(~/.claude/skills/gstack/bin/gstack-repo-mode 2>/dev/null) to load shell configurations.
  • [COMMAND_EXECUTION]: The skill performs extensive command-line operations for environment setup, telemetry, and project verification.
  • Runs various local binaries from ~/.claude/skills/gstack/bin/ during the preamble.
  • Executes xcodebuild and swift build to verify project status after modification.
  • Uses git for branch detection and committing changes to CLAUDE.md.
  • Uses nm to inspect built binary symbols.
  • [DATA_EXFILTRATION]: The skill includes functionality for telemetry and synchronization of project artifacts to external services.
  • Usage telemetry is collected and sent to a remote endpoint via gstack-telemetry-log.
  • Offers synchronization of project artifacts (plans, designs) to a private GitHub repository via gstack-brain-sync.
  • Note: These features are associated with the 'gstack' toolset by the vendor 'garrytan' and are presented as opt-in or configurable.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external project files which constitutes an attack surface for instructions embedded in source code.
  • Ingestion points: The skill reads and globs iOS project source files (.swift) and Package.swift.
  • Boundary markers: The instructions do not specify boundary markers or 'ignore' instructions for the processed content.
  • Capability inventory: The agent has access to Edit, Bash (for compilation and git operations), and AskUserQuestion tools.
  • Sanitization: No explicit sanitization of ingested file content is performed before processing.
  • [PROMPT_INJECTION]: The skill includes model-specific behavioral patches that instruct the agent on its internal reasoning and response style.
  • Instructions for 'todo-list discipline' and 'think before heavy actions' are used to override default agent behaviors.
  • A 'GStack voice' section defines specific prose constraints and vocabulary prohibitions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 18, 2026, 11:46 AM
Security Audit — agent-trust-hub — ios-clean