The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by design. It instructs the agent to recursively search and analyze the existing codebase to identify patterns and constraints.
Ingestion points: SKILL.md (Step 3: Analyze Existing Codebase) directs the agent to use grep and find on project files.
Boundary markers: Absent. There are no instructions to treat file content as data only or to ignore embedded instructions within code comments.
Capability inventory: The skill is granted Write, Edit, Bash, WebSearch, and Read capabilities, which could be exploited if the agent follows malicious instructions found in the codebase.
Sanitization: Absent. The agent processes discovered code directly to inform its architecture designs.
[COMMAND_EXECUTION]: The skill explicitly uses the Bash tool for project exploration.
Evidence includes template commands for grep -r, find, and ls -la targeting the packages/ directory to trace data flows and state management patterns.
[DATA_EXPOSURE]: The skill defines background hooks in its metadata that may log session data or transmit architecture patterns for learning purposes.
after_complete hooks include self-improving-agent and session-logger. While these appear intended for internal performance monitoring, they represent points where technical design details are processed outside of the immediate user interaction.

architecting-solutions