Agent Skills

A collection of reusable skills for AI coding agents. Skills are packaged instructions and helper scripts that extend agent capabilities while keeping the workflow repository-friendly.

Available Skills

handoff-memory

Agent-neutral workflow for creating and maintaining shared repo-local, workspace-wide, or workstream-specific HANDOFF and memory documents.

Use when:

• Writing a project handoff before ending a session
• Writing a workspace handoff from a parent folder that coordinates multiple repos
• Resuming work from an existing handoff
• Standardizing shared project-state notes in Git-trackable files
• Keeping mutable handoff state out of .codex, .claude, .windsurf, or .agents

Behavior:

• Reuses an existing shared handoff file such as docs/HANDOFF.md, memories/HANDOFF.md, or HANDOFF.md
• Defaults to docs/HANDOFF.md for a repo and _memory/HANDOFF.md for a workspace
• Supports workstream-specific workspace documents under _memory/workstreams/<name>/
• Adds helper scripts for create, validate, and staleness checks
• Supports optional timestamped snapshots with explicit kind/reason metadata under docs/handoffs/ or _memory/handoffs/
• Supports global or project-local skill installation, while keeping the shared data inside the repository or workspace root

github-pr-review

Agent-neutral workflow for reviewing GitHub pull requests with gh, local git, tests, and GitHub APIs, then posting review comments as the authenticated GitHub account.

Use when:

• Setting up OAuth or GitHub CLI authentication for PR reviews
• Reviewing a PR URL, owner/repo#123, or the current branch PR
• Leaving review comments as the user's GitHub account
• Reviewing public or private repository PRs
• Collecting PR diff, checks, and related code context before drafting feedback

Behavior:

• Checks gh auth status and confirms the posting account without exposing tokens
• Supports PR URLs, owner/repo#123, PR numbers, branches, and current-branch PR lookup
• Separates public read access from authenticated review posting
• Explains private repo failures such as missing access, org SSO, or insufficient scopes
• Drafts findings first and posts only after confirmation unless immediate posting was requested
• Uses explicit-only approve and request-changes events

github-pr-publish

Agent-neutral workflow for safely publishing GitHub pull requests with gh, local git, and constrained GitHub REST fallback.

Use when:

• Creating, opening, publishing, or preflighting a GitHub pull request from a local branch
• Publishing PRs for public or private repositories through the authenticated GitHub CLI account
• Pushing a local feature branch only as an explicit part of PR creation
• Diagnosing PR creation failures such as missing auth, org SSO, insufficient permission, private repo not-found masking, or validation errors
• Avoiding unsafe gh pr create prompting, fork creation, or accidental pushes

Behavior:

• Previews by default and performs no push, PR creation, browser open, or mutating API call without explicit confirmation flags
• Requires prompt-free PR content and an explicit --head for creation
• Guards pushes behind --push --remote <name> --yes and rejects unsafe branch, fork, force, detached HEAD, and wrong-remote cases
• Supports constrained REST fallback only after remote-head verification
• Includes fake gh/git tests for command construction, no-mutation defaults, and token redaction

commit-helper

Reusable commit-message helper that inspects explicit repo-local rules, recent history, and staged changes before drafting or creating commits.

Use when:

• The user asks for a commit message
• The user wants to commit staged changes but the repo convention is unclear
• A workflow needs to create commits in the target repository's local style
• The current repository may use conventional, gitmoji, plain imperative, or custom commit formats

Behavior:

• Follows explicit local rules > recent history > conservative fallback
• Uses staged changes only when drafting commit messages
• Separates commit format from repo-local phrasing
• Falls back to Conventional Commits when no strong local signal exists
• Supports conventional, gitmoji, plain imperative, and repo-custom styles
• Uses a safe script path for multiline commit bodies without literal \n

Installation

Install from this collection interactively:

npx skills add https://github.com/17-sss/agent-skills

The CLI will inspect the repository, show the available skills, and guide you through the install flow.

Install a specific skill directly:

npx skills add https://github.com/17-sss/agent-skills --skill <skill-name>

Example:

npx skills add https://github.com/17-sss/agent-skills --skill handoff-memory

Usage

Once installed, agents can invoke the relevant skill when a task matches it. Detailed usage, install scope, and workflow notes live inside each skill package under skills/<skill-name>/README.md.

Repository Structure

Each skill lives under skills/<skill-name>/ and may contain:

• SKILL.md - Primary skill definition
• README.md - Human-facing documentation
• AGENTS.md - Agent-facing repo guidance for the skill package
• metadata.json - Catalog metadata
• scripts/ - Helper scripts
• references/ - Supporting docs and templates
• agents/ - Optional agent-specific metadata such as openai.yaml