The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted external codebases and design documents, which creates a surface for indirect prompt injection attacks.
Ingestion points: The workflow in SKILL.md (Phase 1, Step 2) instructs the agent to read all provided documents and scan the target codebase using scripts/scan_codebase.sh. It also searches for repository-local context files such as CONTEXT.md and ADRs.
Boundary markers: The instructions lack explicit delimiters or "ignore embedded instructions" directives to prevent the agent from potentially obeying malicious commands found within the analyzed files.
Capability inventory: The skill possesses the capability to execute local bash scripts and generate detailed reports containing scores and recommendations.
Sanitization: Content read from the external codebase is processed without specific filtering or sanitization measures.
[COMMAND_EXECUTION]: The skill invokes a provided local bash script (scripts/scan_codebase.sh) to perform structural analysis of the target codebase.
Evidence: The instruction in SKILL.md Phase 1, Step 2 explicitly directs the agent to run bash scripts/scan_codebase.sh <codebase_path>.
Risk: This execution pattern relies on the agent platform to properly sanitize the user-supplied path argument to prevent shell command injection.

architecture-reviewer