The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is instructed to locate and execute existing screenshot or recording scripts within the project repository (e.g., using Playwright or trace tools) to gather evidence for UI issues. This behavior involves executing local code found within the analysis context.
[DATA_EXFILTRATION]: The skill performs network operations via GitHub APIs to list, search, and create issues in repositories within the douinc organization. These operations are the primary function of the tool and are controlled by a user confirmation step.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the ingestion of external repository data.
- Ingestion points: The skill reads instructions and configuration from repository files like docs/design-harness.md and processes user-uploaded images/descriptions.
- Boundary markers: The instructions do not define explicit delimiters to separate repository data from the system prompt, nor do they instruct the agent to ignore embedded commands in those files.
- Capability inventory: The agent has the ability to read local files, execute shell commands (for evidence capture), and write to the GitHub API.
- Sanitization: No sanitization or validation logic is specified for the content retrieved from the repository files before it influences the agent's logic.

saylog-uxui-issues