The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to automatically transmit the "Exact error message or stack trace" and file paths to a public GitHub repository. These data structures often inadvertently contain sensitive secrets such as bearer tokens, database connection strings, or internal file system hierarchies that reveal the user's local environment.
[DATA_EXFILTRATION]: The instructions encourage high autonomy by directing the agent to skip user confirmation ("do not stop or wait") and proceed with filing the report immediately. This prevents the user from reviewing or redacting sensitive data before it is published to a public vendor resource.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted error context from the execution environment and passing it directly into a tool with network capabilities.
Ingestion points: Error messages and stack traces encountered during a session (SKILL.md).
Boundary markers: Absent. No instructions are provided to delimit or treat the error message as untrusted content.
Capability inventory: GitHub issue creation (mcp__github__create_issue) which allows external data write.
Sanitization: Absent. There are no guidelines for escaping, filtering, or validating the content of the error messages before they are used to populate the issue template.

auto-bug-reporter