The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes extensive shell command capabilities to perform its refactoring duties.
It uses git for repository state management (status, diff, log, add, commit).
It leverages package managers and build tools including npm, npx, uv, and cargo for code validation.
It automatically executes discovery commands (git rev-parse, find, git status) to establish context upon activation.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because its logic is driven by external data sources.
Ingestion points: The agent reads the REFACTOR_PLAN.md file (which can be manually edited by users or potentially influenced by previous file reads) and the repository's source code files.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard instructions found within the files it reads during the refactoring process.
Capability inventory: The skill possesses high-impact capabilities including Bash execution, file system modification (Write, Edit), and sub-agent delegation via Task.
Sanitization: No specific sanitization or validation logic is defined to prevent malicious instructions embedded in the refactor plan or the source code from being interpreted as valid task steps.

workflow-checkpoint-refactor