launchdarkly-release-plans

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill contains instructions for the agent to use natural language and 'product words' while avoiding the display of internal tool names or the specific skill name to the user. This is categorized as a user experience and branding guideline rather than a malicious attempt to conceal behavior, as the skill simultaneously mandates clear descriptions of all proposed changes and requires user approval for writes.
  • [DATA_EXFILTRATION]: The skill is specifically hardened against credential theft. It strictly forbids the agent from requesting or storing LaunchDarkly API tokens. Instead, it leverages a pattern where the user executes necessary REST commands in their own terminal using local environment variables, ensuring the AI never has access to the sensitive LD_ACCESS_TOKEN.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from LaunchDarkly API responses. While any ingestion of external data theoretically introduces a risk of indirect prompt injection, this skill mitigates the risk by using highly structured data for its operations and implementing a mandatory confirmation step where the user must review the agent's interpreted plan before any action is taken.
  • [COMMAND_EXECUTION]: To handle scenarios where specific tools are missing from the MCP server, the skill provides curl command templates for the user. The instructions clearly state that the agent is not to execute these commands, delegating the execution to the user's local environment to maintain a secure operational perimeter.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 05:22 PM