The Agent Skills Directory

[SAFE]: The skill provides a robust and well-documented framework for evaluating software engineering decisions, focusing on code quality, licensing, and portfolio alignment. It follows a tiered search order that prioritizes internal and trusted resources before external discovery.- [COMMAND_EXECUTION]: The skill utilizes common shell commands such as git clone, mktemp, and rg to perform shallow clones and architectural reviews of repositories. These operations are performed in temporary directories and are restricted to inspection tasks. The instructions include a non-negotiable rule against running arbitrary third-party code.- [DATA_EXPOSURE]: The skill reads local environment configuration files (e.g., ~/.claude/context/manifest.yaml, CLAUDE.md) to understand the user's project structure and provide context-aware recommendations. This data access is localized and necessary for the skill's primary function.- [INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it clones and reads external, untrusted repositories. ● Ingestion points: External Git repositories and package registry manifests. ● Boundary markers: No specific boundary markers are defined for delimiting external file content from agent instructions. ● Capability inventory: The agent has capabilities for shell execution (git, rg), network access (cloning), and file system operations. ● Sanitization: The skill relies on the agent's default handling of file content and does not specify additional sanitization or escaping for ingested data. However, the risk is mitigated by the skill's focus on read-only architectural analysis.

build-vs-clone