The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches configuration and CLI binaries from GitHub's official repository (github.com) and facilitates dependency management through standard registries (NPM, PyPI, Go Proxy, Flutter). These operations target well-known, trusted services and do not escalate the security verdict.
[COMMAND_EXECUTION]: Automates development tasks using the Bash tool for workflows such as artifact building, testing, and deployment. Templates incorporate strict shell settings (set -eu -o pipefail) and overridable variables to ensure execution is predictable and error-resistant.
[DATA_EXFILTRATION]: References sensitive file paths including ~/.private_keys/ for App Store Connect API keys and ~/.local/bin/ for CLI binaries. These references are used exclusively for legitimate operations such as iOS app deployment with official tools (xcrun altool) and binary installation, with no patterns suggesting unauthorized data transmission.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because its primary function involves reading and editing user-provided Makefiles.
Ingestion points: Reads existing project Makefile and related modules during the audit and improvement workflows via the Read and Grep tools.
Boundary markers: Does not implement explicit markers or safety instructions to isolate untrusted Makefile content from agent instructions.
Capability inventory: Authorized to use Bash for execution and Write/Edit for file manipulation.
Sanitization: Content from existing Makefiles is not sanitized before being processed by the agent, which is a common risk for tools designed to refactor existing code.

makefile