The Agent Skills Directory

[Command Execution]: The skill makes extensive use of local command-line tools including git, gh, jq, and python3. These tools are used to resolve PR metadata, fetch code, and post results back to GitHub. This behavior is necessary for the tool's core functionality of repository analysis and PR integration.
[Dynamic Execution]: The skill utilizes Python heredocs within shell scripts (e.g., in coverage-report.sh and prepare-review-context.sh) to generate and execute processing logic at runtime. Additionally, the subagent prompts in references/prompts.md mention the creation of temporary analysis scripts. These patterns are used for data transformation and verification tasks during the review process.
[Indirect Prompt Injection]: The skill ingests untrusted data from external sources, specifically PR diffs and existing GitHub review threads. This data is processed by subagents, creating a potential surface for indirect prompt injection. The skill includes mitigation strategies such as instructing subagents to read specific context files in a defined order and using explicit prompt templates to bound the analysis.

doris-repo-review