The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents vulnerability surfaces for indirect prompt injection.
Ingestion points: User inputs regarding technology stack, infrastructure, and budget constraints are ingested through trigger phrases in SKILL.md and interpolated into generated artifacts.
Boundary markers: The skill does not use delimiters or "ignore instructions" warnings when processing these inputs to separate user data from system instructions.
Capability inventory: The skill performs file writes (configuration generation) and documents shell command execution (installation scripts).
Sanitization: No sanitization or validation of user-provided content is performed before interpolation into configuration artifacts.
[PROMPT_INJECTION]: Multiple reference files (e.g., references/architecture-patterns.md, references/state-management.md, references/testing-strategies.md) contain 150 repetitive sections of text. This extreme volume of redundant content (over 100,000 words across files) poses a risk of context displacement, potentially causing the agent to lose track of its safety constraints or system instructions during runtime processing.
[EXTERNAL_DOWNLOADS]: The skill documents the retrieval and execution of setup scripts for the Datadog Agent and New Relic CLI from their official domains (s3.amazonaws.com/dd-agent and download.newrelic.com). These are established services and the downloads are for their intended primary purpose.
[COMMAND_EXECUTION]: Provides instructions for executing shell commands and editing system configuration files (e.g., vim /etc/datadog-agent/datadog.yaml) related to the setup of observability tools. These commands are typical for the skill's stated purpose and focused on tool-specific directories.

apm-observability