The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the github/agent-task extension from GitHub's official repository via the GitHub CLI. As this is a well-known service and the resource belongs to the skill's authoring organization, the reference is considered safe.
[CREDENTIALS_UNSAFE]: The skill documentation correctly identifies the need for Personal Access Tokens (PATs) and recommends storing them as repository secrets (e.g., COPILOT_GITHUB_TOKEN). It discourages hardcoding credentials and provides clear guidance on secure token management.
[PROMPT_INJECTION]: The skill describes a workflow that ingests natural language instructions (e.g., from issue descriptions or files) to trigger automated code modifications via GitHub Copilot.
Ingestion points: Instructions enter the system via the gh agent-task create command arguments or the --from-file parameter.
Boundary markers: The documentation does not specify explicit boundary markers or sanitization for these instructions.
Capability inventory: The tool can create issues and pull requests that result in automated code changes.
Sanitization: No specific sanitization methods for input instructions are described. While this represents a surface for indirect prompt injection, it is the intended functionality of the tool and follows standard agentic design patterns.

gh-agent-session