The Agent Skills Directory

Command Execution (Subprocess Calls): The skill instructs the agent to execute shell commands to gather logs and submit issues. While command execution is a sensitive capability, this skill implements authentication checks and requires explicit user confirmation before any action is taken.
Evidence: Use of gh issue create --repo microsoft/azure-sql-database-container in SKILL.md to submit reports.
Evidence: Use of docker logs sqldb in SKILL.md to gather diagnostic facts for the report.
Data Redaction and Privacy: The skill contains explicit, mandatory instructions to identify and remove sensitive credentials from reports before they are sent to a public repository. This practice is an essential safeguard for maintaining user privacy and security.
Evidence: Step 3 (Redaction) in SKILL.md lists MSSQL_SA_PASSWORD, docker login credentials, and access tokens for removal.
Evidence: Instructions in references/issue-fields.md to replace connection string secrets with ***.
Indirect Prompt Injection Surface: The skill processes external data, such as user prompts and interaction history, to populate bug report templates. This constitutes an attack surface where instructions could potentially be embedded in the report body.
Ingestion points: User prompts, skill instructions, and container logs (SKILL.md).
Boundary markers: The skill mandates that the agent show the user the full title and body first and wait for explicit confirmation before submission.
Capability inventory: The skill uses gh issue create and docker logs (SKILL.md).
Sanitization: The skill implements strict redaction of common secret patterns and identifiers.

azuresql-db-feedback