The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs subprocess.run within several Python scripts (arch_drift.py, validate_index.py, verify_arch_plan.py) to interact with the local git CLI. These commands are used for structural analysis, such as listing tracked files, diffing commits, and verifying repository state. The calls are implemented using list-based arguments without shell=True, which is a safe practice to prevent shell injection.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository, creating a potential surface for indirect prompt injection.
Ingestion points: The agent reads architectural documentation files (e.g., ARCHITECTURE.md, STACK.md) and crawls the repository file tree.
Boundary markers: Not explicitly defined for input file contents; however, the skill's logic is primarily focused on structural metadata rather than executing code found in prose.
Capability inventory: The skill performs local file read/write operations and executes git commands via subprocess.
Sanitization: The verify_arch_plan.py script validates that all intended paths are repository-relative and do not contain .. directory traversal sequences, preventing escapes.- [DATA_EXPOSURE]: A security-focused script, repo_tree.py, implements a comprehensive deny-list that prevents the agent from reading sensitive files or directories. This includes credentials (.env), version control internals (.git), and development environment configurations (.ssh, .vscode, node_modules). This proactive filtering reduces the risk of sensitive data exposure during codebase mapping.

arch-index