The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses extensive shell commands via git and gh (GitHub CLI) to manage branch topology, PR metadata, and remote synchronization. This includes complex operations like git rebase, git push --force-with-lease, and GitHub API interactions (gh api).
[COMMAND_EXECUTION]: For its own internal development, the skill executes local Python scripts (scripts/validate_evals.py, scripts/generate_manifest.py, and scripts/evaluate_package.py) using uv run. These are used for validation and manifest generation.
[SAFE]: The skill implements robust safety checks and 'Stop Conditions' across all workflows. It mandates git status --porcelain checks to prevent operations on dirty worktrees and uses --force-with-lease to prevent overwriting remote changes that haven't been fetched.
[SAFE]: The skill contains specific logic to prevent 'Indirect Prompt Injection' or topology corruption by prioritizing explicit user-supplied branch order and PR metadata over ambiguous automated inference.
[SAFE]: Instructions explicitly restrict the execution of internal armory development scripts to the skill's own repository, preventing the agent from running these scripts on arbitrary user projects.

stacked-prs