The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute database client commands (psql, mysql) that rely on the $DATABASE_URL environment variable. This variable frequently contains sensitive plaintext credentials, including usernames and passwords, which are exposed during the execution of profiling queries.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes external, untrusted content from the local codebase and database engine metadata, presenting a surface for indirect prompt injection.
Ingestion points: The skill ingests untrusted data via the Read and Grep tools while scanning source files for anti-patterns (Step 3 and 4) and via the Bash tool when capturing output from database performance schemas (Step 2).
Boundary markers: Absent. The skill does not employ delimiters or guardrails to distinguish between its own instructions and the content of the files or database records it analyzes.
Capability inventory: The skill is authorized to use the Bash tool, which provides the ability to execute shell commands based on the findings of the performance audit.
Sanitization: There is no implemented sanitization or validation of the ingested code snippets or database query strings before they are incorporated into the performance report or used to suggest code modifications.

perf-profile