The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill teaches a workflow that involves the ingestion and analysis of external data which could potentially contain malicious instructions or malformed data designed to influence agent behavior.
Ingestion points: The skill processes various performance reports (e.g., k6 summary.json, Locust results_stats.csv), database query plans (plan.json), and system CPU profiles in the incident triage workflow and associated evaluation tasks.
Capability inventory: The agent is instructed to use powerful tools such as k6, jq, node, python, and various system profilers (py-spy, async-profiler, clinic.js) to evaluate this data.
Boundary markers: There are no explicit markers or 'ignore embedded instructions' warnings provided for the data being parsed, which is a common defense against indirect injection.
Sanitization: The skill does not describe any sanitization or validation of input data, such as target endpoints or query strings, before using them in diagnostic commands or triage reports.
[COMMAND_EXECUTION]: The skill instructs the agent to perform diagnostic operations that involve executing shell commands and system profilers.
The triage workflow in SKILL.md directs the agent to run profilers like py-spy, async-profiler, and Go pprof to capture flame graphs, which typically require elevated system privileges to attach to running processes.
The skill includes several examples of using jq to parse data and running various load testing binaries and build tools (mvn, gradlew, locust).
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and installing performance testing software from official and well-known technology sources.
SKILL.md contains installation commands for k6 using established package managers like brew, Chocolatey, and winget, as well as official Docker images.
references/locust.md and references/gatling.md describe installing locust via pip and Gatling plugins via standard Java build tools. These downloads target official project domains and registries (e.g., Apache, Grafana, Locust.io).

load-testing-overview