The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash(curl:*), Bash(jq:*), and Bash(date:*) to perform its primary function: querying observability backends. All network operations are directed at endpoints defined by the user in environment variables (VM_METRICS_URL, VM_LOGS_URL, etc.). This is the intended behavior for an investigation skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as logs and traces.
Ingestion points: Data is fetched from VictoriaLogs, VictoriaMetrics, and VictoriaTraces backends via curl (defined in SKILL.md and all subagents).
Boundary markers: The skill emphasizes a structured "Investigation Protocol" and uses subagents to isolate data gathering from interpretation, which provides a natural boundary.
Capability inventory: The skill has access to shell tools (curl, jq, date), file reading, and agent dispatching.
Sanitization: Results are parsed using jq to extract specific fields, reducing the likelihood of raw malicious content influencing the agent's core logic.
[CREDENTIALS_UNSAFE]: The skill implements safe secret management. It uses the VM_CURL_CONFIG environment variable to point to a local file containing authentication headers, explicitly instructing the agent to never print the contents of this file. This avoids exposing credentials in process lists or logs.

investigating-with-observability