The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes shell commands including az rest, git show, and grep to interact with the local repository and the Azure DevOps API. This is a primary function for retrieving code and reporting findings.
[EXTERNAL_DOWNLOADS]: Fetches pull request information and diffs from Azure DevOps (dev.azure.com). This is a well-known service and the activity is part of the core code review workflow.
[DATA_EXFILTRATION]: Transmits code review findings to Azure DevOps PR comments. While this involves sending analysis of local source code to an external service, it targets a well-known platform as part of the user-initiated review process.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests and processes untrusted data from code diffs.
Ingestion points: Untrusted data enters the agent context via the Review target in Step 1, which includes PR diffs or repository files.
Boundary markers: The skill uses XML-style tags like <review_context> to structure information, but lacks explicit warnings to ignore instructions embedded in the code being reviewed.
Capability inventory: The agent has the ability to execute shell commands and perform network operations via the az CLI.
Sanitization: No explicit sanitization or filtering is performed on the ingested code content before analysis.

deep-code-review