The Agent Skills Directory

[SAFE]: The skill acts as a comprehensive reference for the Qodo Merge product. All instructions and examples follow standard development practices and do not contain malicious instructions or hidden code.
[EXTERNAL_DOWNLOADS]: The documentation references official vendor resources, including Docker images from 'codiumai/pr-agent' and GitHub repositories under 'qodo-ai'. These are legitimate components of the documented service.
[COMMAND_EXECUTION]: The skill provides CLI command examples and configuration snippets for users to set up the PR-Agent tool in their own environments. These commands are standard for DevOps tooling.
[DATA_EXFILTRATION]: The skill discusses the requirement for various API tokens (OpenAI, GitHub, Jira, etc.) but correctly instructs users to manage these via secure environment variables or CI/CD secrets rather than hardcoding them.
[PROMPT_INJECTION]: The skill describes a tool that processes untrusted data from pull requests, which presents a surface for indirect prompt injection. Ingestion points: The tool reads pull request diffs, descriptions, and comments across GitHub, GitLab, and Bitbucket (referenced in references/tools.md). Boundary markers: The documentation does not specify internal prompt delimiters but mentions preserving user descriptions. Capability inventory: The tool can perform automated code reviews, generate suggestions, post comments, and interact with Jira (referenced in references/core-abilities.md). Sanitization: Documentation advises that AI errors are rare but possible and recommends human review and the use of extra_instructions to guide model behavior.

qodo-merge