The Agent Skills Directory

[SAFE]: The skill implements a secure-by-design workflow for issue management, explicitly prohibiting the agent from merging its own pull requests or taking work that has not been human-verified.
[COMMAND_EXECUTION]: Uses standard development tools including git, gh (GitHub CLI), and jq. Shell scripts are designed to safely handle external data (such as issue titles) by using jq for sanitization before using them as command-line arguments, which mitigates command injection risks.
[DATA_EXFILTRATION]: No data exfiltration detected. The skill's network activity is confined to interacting with the official GitHub API and local repository operations.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it processes untrusted data from GitHub issue bodies. Ingestion points: Step 1 (Read the frontier) and Step 3 (Claim it) in SKILL.md fetch issue lists and details. Boundary markers: The instructions do not define specific delimiters for separating issue body content from the agent prompt. Capability inventory: The agent can modify issues and create pull requests via gh, and modify the codebase via git. Sanitization: The agent is instructed to focus on acceptance criteria, but no technical sanitization is applied to the issue body text itself. This risk is inherent to the tool's primary purpose and is managed by defined workflow constraints.

work-next-issue