The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes application performance data, slow query logs, and profiling metrics. This establishes an attack surface where maliciously crafted application output could potentially influence the agent's analysis.
Ingestion points: Performance metrics, database query logs, and bundle analysis results from processed code and logs.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are provided for the data ingested during analysis.
Capability inventory: Extensive use of CLI tools (clinic, py-spy, k6), file system access for writing snapshots, and network requests for load testing.
Sanitization: No explicit sanitization or validation of the ingested performance data is mentioned in the procedures.
[DATA_EXFILTRATION]: The skill provides code to generate heap snapshots via v8.writeHeapSnapshot(). Heap snapshots contain the full memory state of the application, which may include sensitive information like session tokens, credentials, or user data. The skill correctly recommends protecting this diagnostic endpoint with authentication.
[DYNAMIC_EXECUTION]: The skill generates specialized profiling scripts for Node.js (scripts/memory-profile.mjs) and Python (scripts/profile_endpoint.py) using predefined templates. This is a standard and expected practice for performance diagnostics.
[COMMAND_EXECUTION]: Instructions include the installation and use of several powerful performance analysis tools such as clinic.js, py-spy, pprof, and k6, which require shell command execution capabilities.

performance-profiler