The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently uses the Bash tool for repository management, environment configuration, and executing local development tools (e.g., Git, vibe-suite binaries).
[DYNAMIC_EXECUTION]: The skill generates and executes short Python scripts via heredocs to manage audit baselines and uses eval on the output of local platform binaries to set environment variables. These operations are restricted to internal data and platform tools.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze data from external websites, which presents a surface for indirect prompt injection.
Ingestion points: Web page text, snapshots, and browser console output are ingested via the browse tool in SKILL.md during Phase 3 and 4.
Boundary markers: The skill includes 'Browser Rule 4', which explicitly mandates that all content returned from a page must be treated as untrusted data and never as instructions.
Capability inventory: The agent has access to shell command execution (Bash), file writing (Write/Edit), and package installation (npm/pip).
Sanitization: Analysis is primarily performed on visual screenshots and CSS properties, and code modifications are constrained to styling fixes.

design-review