Agent Skills

Reusable skills for software-engineering agents.

Available skills

to-clean-pr

Drive a feature, bugfix, or existing GitHub PR or GitLab MR through an evidence-backed loop: one isolated head, verified changes, valid review fixes, and two clean polls.

delegate-to

Delegate any task to an external agent CLI (/delegate-to cursor, /delegate-to hermes). Write tasks run in the backend's own git worktree, in the background, under the backend's own approvals; the calling agent verifies the result and resumes the same session for follow-ups. Add a backend by adding backends/<name>.md.

consult

User-invoked, read-only second opinion from another agent (/consult <backend>) for contested design, debugging, architecture, security, or trade-off questions. Runs through delegate-to's read-only mode with evidence and rebuttal rounds in one session; install it together with delegate-to.

humanize

Improve readability while preserving facts, uncertainty and the requested voice. Includes calibration examples and a linter for style and character-level review.

squash-rebase

User-invoked squash of a branch's unique commits onto its open GitLab MR's target branch, one commit per consecutive same-author run with that author kept, pushed with force-with-lease. Auto-detects commits a stacked lower MR already re-pushed, and rewrites only in a temporary worktree.

Install

This repository has not been published yet. After its first push, list the available skills:

npx skills add alvedder/skills --list

Install a named skill:

npx skills add alvedder/skills --skill <skill-name>

Repository structure

skills/
	├── consult/
	│		├── SKILL.md
	│		└── agents/
	│				└── openai.yaml
	├── delegate-to/
	│		├── SKILL.md
	│		├── backends/
	│		│		├── cursor.md
	│		│		└── hermes.md
	│		└── agents/
	│				└── openai.yaml
	├── humanize/
	│		├── SKILL.md
	│		├── PATTERNS.md
	│		├── EXAMPLES.md
	│		├── LINT.md
	│		└── scripts/
	├── squash-rebase/
	│		├── SKILL.md
	│		└── agents/
	│				└── openai.yaml
	└── to-clean-pr/
			├── SKILL.md
			├── references/
			└── agents/
					└── openai.yaml

Each skill is self-contained. SKILL.md is the portable skill definition; agents/openai.yaml adds Codex-facing display metadata.

Development

Keep the skill folder name and frontmatter name aligned. Before publishing changes, validate local discovery:

npx skills add . --list

License

MIT