The Agent Skills Directory

[COMMAND_EXECUTION]: Employs read-only codebase scanning using patterns like grep and glob to identify orphaned components and unwired event handlers in files such as references/detection-taxonomy.md. These operations do not involve executing shell scripts or modifying system state.
[EXTERNAL_DOWNLOADS]: References the use of Firecrawl and Playwright via the Model Context Protocol (MCP) to verify current industry standards and runtime UI behavior. These tools are managed by the host platform and are not used to download or execute arbitrary external payloads.
[DATA_EXFILTRATION]: Audits configuration files like .env.example and package.json to ensure integrations like Supabase and Sentry are correctly initialized. The skill's preservation contract explicitly forbids the fabrication or unauthorized transmission of environment keys or DSNs.
[PROMPT_INJECTION]: The skill instructions in SKILL.md and references/preservation-contract.md reinforce behavioral constraints and safety boundaries. There are no attempts to override agent safety protocols or bypass user approval for proposed changes.
[SAFE]: The skill represents a legitimate developer utility for codebase maintenance. It ingestion point involves local repository files, while its boundaries are defined by a multi-pass methodology that requires specific file citations and human review of all identified issues. Capabilities are limited to auditing and planning as defined in references/detection-methodology.md.

plan-stub-checker