The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions repeatedly use npx @leoflores/datadog-cli to download and execute code from a third-party npm package at runtime. This package is managed by an individual user and is not an official Datadog tool or from a recognized trusted vendor.
[INDIRECT_PROMPT_INJECTION]: The skill provides tools to ingest and process untrusted external data from Datadog, which could be influenced by an attacker.
Ingestion points: Commands like logs search, logs tail, and dashboards get (referenced in SKILL.md and references/dashboards.md) pull data from Datadog into the agent's environment.
Boundary markers: None. The instructions do not provide delimiters or instructions to ignore potential commands embedded within log entries or dashboard metadata.
Capability inventory: The agent has the capability to execute shell commands (npx), write to the local filesystem (--output flags, /tmp/dashboard.json), and perform network operations via the Datadog API.
Sanitization: There is no evidence of sanitization or validation of the content retrieved from Datadog before it is used in subsequent instructions or shell interpolation.
[COMMAND_EXECUTION]: The workflow described in references/dashboards.md relies on shell variable interpolation (e.g., "$DESCRIPTION", "$WIDGETS") to pass data retrieved from a JSON file back into a CLI command. If the retrieved data contains malicious payloads and the shell environment or the CLI tool handles these strings unsafely, it could lead to command injection.

datadog-cli