The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently uses python3 -c to execute Python code snippets for JSON parsing, environment switching, and log processing. This pattern involves interpolating shell-derived variables (like ENV_KEY, API_PATH, and TOKEN) into executable strings, which can be vulnerable to command injection if inputs are not strictly validated.
[CREDENTIALS_UNSAFE]: The skill instructions manage "Grafana Service Account Tokens" and provide shell commands to write these tokens into a local JSON file (environments.json). Storing and handling plaintext secrets in skill-specific configuration files increases the risk of credential exposure.
[DATA_EXFILTRATION]: The skill uses curl to transmit authentication tokens to external domains (e.g., test13.xnzn.net, monitor-dev.xnzn.net). These domains are not recognized as trusted organizations or well-known services. Sending credentials to unknown external endpoints is a sensitive operation that requires caution.
[PROMPT_INJECTION]: The skill processes arbitrary log data from an external Loki instance. This represents an indirect prompt injection surface as logs can contain data from untrusted users.
Ingestion points: API response data from curl calls to Loki query_range endpoints (found in SKILL.md).
Boundary markers: Absent; log lines are processed directly without delimiters or instruction-bypass warnings.
Capability inventory: The skill possesses network access (curl), file-write capabilities (updating environments.json), and dynamic execution (python3 -c).
Sanitization: Absent; the skill uses basic string splitting (parts = line.split(',')) and regex without validation of content safety.

loki-log-query