The Agent Skills Directory

[PROMPT_INJECTION]: The skill performs indirect processing of untrusted data by reading and analyzing front-end source files (TSX, Vue, Svelte, Angular, CSS). While this creates a potential surface for indirect prompt injection where malicious comments could attempt to influence the agent's logic, it is the primary function of the skill. The ingestion involves sampling up to 40 representative files to detect project dialects, and while no specific boundary markers are defined for this read process, the skill requires explicit user approval before applying any renames, providing a human-in-the-loop safety check.\n- [COMMAND_EXECUTION]: The skill utilizes command-line tools to verify project integrity. Specifically, it executes grep to find identifiers, tsc --noEmit for type checking, and vitest run for testing. Additionally, it runs local lifecycle and orchestration scripts (e.g., checklist.mjs, orchestrate.mjs) located within the skill's installation directory (~/.claude/skills/refactor-chain/scripts/). These operations are consistent with the tool's stated purpose as a refactoring utility.

refactor-web-05-naming