Agent Skills for Software Engineering

Reusable Agent Skills for practical software-engineering work.

Install only the capabilities you need, or use the complete catalog. Each Skill owns one clear kind of work—such as repository mapping, implementation, review, browser operation, or external-AI collaboration—and keeps its authority boundary explicit.

Quick Start

Browse the 17 available Skills:

npx skills@latest add idaibin/skills --list

Choose Skills interactively:

npx skills@latest add idaibin/skills

Install selected Skills globally for Codex:

npx skills@latest add idaibin/skills \
  --skill repo-map repo-review ask-ai \
  --global --agent codex

Install the complete catalog:

npx skills@latest add idaibin/skills \
  --skill '*' --global --agent codex --yes

See INSTALL.md for project installs, suggested sets, updates, removal, and other supported agents.

Catalog

Skill What it helps with
repo-map Map repository roots, architecture, commands, dependencies, ownership, and reusable contracts.
domain-modeling Resolve shared business terms, rules, lifecycles, and domain boundaries.
product-spec Turn product decisions into implementation-ready behavior, states, and acceptance criteria.
ui-spec Turn an accepted visual source into a traceable UI contract.
dev-frontend Implement and validate frontend features, refactors, tooling, and selected-source UI work.
dev-java Implement Java and Spring changes against the repository's real build and runtime contracts.
dev-rust Implement Rust features and refactors with ownership, safety, and behavior evidence.
audit-frontend Audit a bounded frontend surface without modifying it.
audit-java Audit Java and Spring architecture, security, transactions, persistence, and integration.
audit-rust Audit Rust ownership, concurrency, persistence, performance, memory, and unsafe boundaries.
repo-review Review a Worktree, commit, range, or verified package on a fixed evidence basis.
repo-delivery Commit, integrate, push, and clean up reviewed changes with explicit Git authorization.
ops-browser Operate and verify browser pages while preserving target and evidence boundaries.
ops-client Operate and verify real desktop-client processes and windows.
ask-ai Prepare or run explicitly authorized external-AI review, research, relay, and image workflows.
workspace-taskboard Show and coordinate Codex task cards within one verified local project root.
human-writing Draft, rewrite, diagnose, and adapt source-grounded writing without changing its facts or voice.

How They Work Together

Start with the Skill closest to the requested outcome. Add another owner only when the task genuinely crosses an authority boundary.

unknown repository -> repo-map
unclear product     -> domain-modeling / product-spec
UI contract         -> ui-spec -> dev-frontend
source change       -> matching dev-* owner
bounded audit       -> matching audit-* owner
change review       -> repo-review
Git delivery        -> repo-delivery
browser/client proof -> ops-browser / ops-client
external AI         -> ask-ai
project task board  -> workspace-taskboard

Handoffs transfer scoped evidence, not permission. For example, implementation does not authorize a commit, and an external review does not authorize source changes. Every package is self-contained at runtime and can be installed independently.

Find the Right Skill

From a source checkout, search by task, stack, capability, or boundary:

python3 scripts/search-skills.py "review release risk"
python3 scripts/search-skills.py "implement a Java migration" --json

The search uses skills-index.json. Runtime discovery continues to use each package's portable SKILL.md metadata.

Documentation

ask-chatgpt was renamed to ask-ai; legacy wording remains compatible, but ask-ai is the maintained package.

Development

Run the repository's complete validation entry point before publishing changes:

bash scripts/check-skills.sh

Detailed validation rules live in skills/AGENTS.md.

License

Licensed under the Apache License 2.0.