The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to locate and read sensitive configuration files, such as .claude/settings.json, .vscode/launch.json, and .gemini/settings.json. These files frequently contain API keys, authentication tokens, and environment variables which could be exposed to the model.
[COMMAND_EXECUTION]: The diagnostic process involves running shell commands that incorporate user-supplied input. Specifically, the agent is instructed to run gh issue list using an unsanitized error snippet in the search query, which could be exploited for command injection if the error message is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes documentation from a GitHub repository and search results from the GitHub CLI, creating a surface for indirect prompt injection.
Ingestion points: Reads content from https://github.com/ChromeDevTools/chrome-devtools-mcp/blob/main/docs/troubleshooting.md and output from gh issue list searches.
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore embedded commands within the fetched external data.
Capability inventory: The skill possesses file reading capabilities and the ability to execute shell commands via npx and gh.
Sanitization: There is no evidence of sanitization or validation performed on the external content before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill fetches troubleshooting guidelines from the official ChromeDevTools repository on GitHub.
[DYNAMIC_EXECUTION]: The agent uses npx to execute the chrome-devtools-mcp package directly from the npm registry for environment verification and logging.
[REMOTE_CODE_EXECUTION]: The diagnostic steps include downloading and running code from the npm registry using npx chrome-devtools-mcp@latest.

troubleshooting