The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data, including pull request descriptions and code diffs. This configuration creates a vulnerability surface for indirect prompt injection, where an attacker could embed malicious instructions within a code repository to manipulate the agent's review output or behavior.
Ingestion points: The agent processes $ARGUMENTS, pr_description, and code_diff within the review_prompt template.
Boundary markers: The skill uses markdown headers (e.g., **Change Summary:**) to separate sections in its internal prompt but does not include explicit instructions for the AI to ignore embedded commands within the untrusted input.
Capability inventory: The skill possesses the capability to generate review comments and interact with external APIs (GitHub API via scripts) to post findings.
Sanitization: The provided templates do not demonstrate explicit sanitization or validation of the input data before it is interpolated into the LLM context.
[COMMAND_EXECUTION]: The provided automation templates and Python scripts utilize system-level commands to execute static analysis tools.
Evidence: The Python CodeReviewOrchestrator class uses subprocess.run to call sonar-scanner and semgrep.
Context: These executions are standard for the skill's primary purpose of automated code analysis and are limited to specific, well-known security binaries.
[EXTERNAL_DOWNLOADS]: The skill's GitHub Action templates reference official actions from GitHub's marketplace.
Evidence: Usage of actions/checkout@v4 and actions/github-script@v7.
Context: These are well-known, trusted components within CI/CD ecosystems.

code-review-ai-ai-review