The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository-controlled files to determine workflow steps and PR content.
Ingestion points: Reads AGENTS.md, CONTRIBUTING* documents, and pull_request_template.md from the repository file system.
Boundary markers: Absent; the skill does not use specific delimiters or instructions to ignore potential instructions embedded within these external files.
Capability inventory: Executes git and gh shell commands, writes to the local file system (creating worktrees and branches), and performs network operations (pushing code and creating PRs).
Sanitization: None; the skill is instructed to strictly follow the templates and guidelines found in the repository without filtering content.
[EXTERNAL_DOWNLOADS]: The skill references external documentation for the gh stack extension to guide its operations.
Evidence: Includes multiple links to official documentation hosted at github.github.com/gh-stack/* for overview, quick start, and CLI reference.
Note: These resources originate from a well-known service and are documented neutrally for transparency.
[COMMAND_EXECUTION]: The skill performs extensive shell command execution using the git and gh CLI tools to manage the development environment.
Evidence: Dynamically discovers syntax via commands like git --help, git worktree --help, and gh stack --help to determine available flags and subcommands.
Mitigation: The skill includes a "Primary checkout publish guard" to prevent unintended branch switching and requires explicit user opt-in for worktree creation and destructive operations like merging or force-pushing.

git-workstreams