The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes system commands via the bash tool to perform build checks and run test suites, specifically using npm test, bun test, pytest, tsc --noEmit, and eslint .. These executions are integral to the skill's verification phases to ensure that refactoring does not introduce regressions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of ingesting and processing external codebase files and user-supplied arguments.
Ingestion points: User input provided via $ARGUMENTS and local source code files accessed through exploration sub-agents and Language Server Protocol (LSP) tools.
Boundary markers: The prompt utilizes a structured, phase-based template (e.g., <user-request> tags) and an 'Intent Gate' to validate goals, though it does not explicitly instruct the agent to ignore potentially malicious instructions embedded in the code comments of the files being refactored.
Capability inventory: The skill possesses significant capabilities, including the ability to write to the filesystem (edit, lsp_rename, ast_grep_replace) and execute arbitrary shell commands (bash).
Sanitization: The skill employs several defensive layers, such as mandatory verification checkpoints after every modification, a dedicated 'Plan' review phase, and a 'Final Verification' phase to identify and stop if errors or unexpected behaviors occur.

refactor