The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It scans project files (e.g., lib/**/*.ex) and passes their content or context to downstream analysis agents (phx:elixir-reviewer, phx:otp-advisor). Malicious instructions embedded within application code comments or metadata could potentially influence the behavior of these secondary agents.
Ingestion points: Project file scanning via find lib/ -name "*.ex" and subsequent reading of module/context paths.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are visible in the prompts sent to spawned sub-agents.
Capability inventory: The skill can execute shell commands, spawn further agents, and execute arbitrary Elixir and SQL code through the Tidewave MCP.
Sanitization: There is no evidence of sanitization or filtering applied to the content of the project files before they are processed by the LLM sub-agents.
[COMMAND_EXECUTION]: The skill leverages the Tidewave MCP (mcp__tidewave__project_eval, mcp__tidewave__execute_sql_query) to execute Elixir code and SQL queries against the running environment.
Evidence: Usage of Repo.query!("EXPLAIN ANALYZE ...") and Process.info(pid, :message_queue_len) are used to profile the system. While these are high-privilege operations, they are integral to the skill's primary function of performance analysis.

perf