The Agent Skills Directory

[PROMPT_INJECTION]: The instructions focus on task classification and execution routing. There are no patterns suggesting safety filter bypasses, instruction overrides, or system prompt extraction attempts.
[DATA_EXFILTRATION]: The skill does not access sensitive directories (e.g., .ssh, .aws) or perform unauthorized network operations. Use of tools is limited to the local codebase context.
[COMMAND_EXECUTION]: Shell access via the Bash tool is explicitly restricted to validation tasks such as typechecking, running tests, or building the project, which is standard behavior for development-oriented skills.
[EXTERNAL_DOWNLOADS]: No external dependencies, remote scripts, or third-party package managers are utilized in the skill definition.
[OBFUSCATION]: The content in both SKILL.md and SKILL.ko.md is presented in clear text. No Base64 encoding, zero-width characters, or hidden URL patterns were identified.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an execution engine that processes untrusted codebase data.
Ingestion points: Local codebase files accessed via Read/Grep/Glob and user task descriptions in SKILL.md.
Boundary markers: The skill requires the use of sequential-thinking to plan actions before execution, which serves as a cognitive boundary.
Capability inventory: File writing (Edit/Write) and shell command execution (Bash) are used for implementation and verification in SKILL.md.
Sanitization: The skill relies on agent-led validation (test/build) rather than explicit data sanitization filters.

execute