The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill processes untrusted external data from REST API responses and HTTP headers without explicit sanitization or boundary markers.
Ingestion points: Untrusted data enters the agent context via curl output, REST API response headers (x-qm-*), and _envelope responses during performance inspection.
Boundary markers: The instructions do not define delimiters or specific 'ignore embedded instruction' warnings for the data being analyzed.
Capability inventory: The skill possesses the capability to execute a local Node.js script (perf_inspect.mjs) and WP-CLI commands including arbitrary PHP code execution via wp profile eval.
Sanitization: No sanitization, escaping, or validation of the external content is described before the agent processes it.
[Dynamic Execution] (LOW): The skill utilizes dynamic code execution as part of its core profiling workflow.
Evidence: Step 3 recommends using wp profile eval for targeted code path analysis, which executes a provided PHP string.
Context: Although this is a high-privilege capability, it is directly related to the skill's primary purpose (WordPress performance profiling), justifying a downgrade in severity from MEDIUM to LOW.
[Metadata Poisoning] (LOW): The skill contains misleading metadata by referencing a future WordPress version (6.9) and a documentation URL dated November 2025, both of which are currently non-existent. While not containing malicious instructions, this represents deceptive documentation content.

wp-performance