The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Bash scripts (agent-rules.sh, create-symlinks.sh, and verify-architecture.sh) to manage the repository environment. These scripts perform standard file system operations, including ln -sf for symbolic link creation and mkdir for directory initialization, to align agent configurations with a central source.- [PROMPT_INJECTION]: The centralization of instructions into AGENTS.md and the .agents/skills/ directory creates a surface for indirect prompt injection. Ingestion points: All configured agents ingest instructions from the centralized AGENTS.md file and tools from the .agents/skills/ directory via symlinks. Boundary markers: None are employed; content from the central source is linked directly into the agent's primary instruction files (e.g., CLAUDE.md). Capability inventory: The skill possesses Read, Write, and Bash capabilities for environment management, while the agents consuming the instructions often have broad access to project code and execution tools. Sanitization: No validation or sanitization of the content in the central source files is performed by the skill.- [SAFE]: No malicious patterns such as data exfiltration, hardcoded credentials, remote code execution from untrusted sources, or obfuscation were identified. All script behaviors are transparent and consistent with the documented goal of repository-wide agent alignment.

agent-rules