The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to generate and process "heap dumps for memory analysis". Heap dumps capture the entire memory state of an application and frequently contain sensitive information, including authentication tokens, passwords, and private user data that were resident in memory at the time of the dump.
[DATA_EXFILTRATION]: Instructions include analyzing database "slow query logs" and "database query profiling". These logs can expose sensitive data values included in queries or leak information about the database schema and data distribution.
[COMMAND_EXECUTION]: The skill empowers the agent to execute significant changes to the environment, including modifying database structures (indexes), application source code (algorithms, caching), and infrastructure configurations (CDN rules, CI/CD pipelines).
[PROMPT_INJECTION]: The workflow creates an indirect prompt injection attack surface by accepting untrusted input through the $ARGUMENTS variable and various system logs, which are then passed to specialized subagents without adequate sanitization or boundary markers.
Ingestion points: User-supplied $ARGUMENTS and various diagnostic data sources (logs, traces, profiling output).
Boundary markers: None. The skill lacks delimiters or "ignore" instructions to prevent the agent from following commands embedded in the ingested data.
Capability inventory: Modifications to database schemas, application source code, and CI/CD configurations.
Sanitization: There is no evidence of input validation or content sanitization before the data is processed by the LLM.
[EXTERNAL_DOWNLOADS]: The skill refers to several well-known services such as DataDog, New Relic, Cloudflare, and CloudFront for observability and content delivery optimization.

application-performance-performance-optimization