The agent skills I use for engineering, product, research, writing, and everyday tools.
A typical engineering path, with skills chosen as needed. I clarify the goal and break down larger tasks before implementing, reviewing, simplifying, and publishing. Dashed arrows show optional paths; I fix findings and rerun relevant checks before publishing.
flowchart TD
G["grill-me · clarify the goal"] -.-> W["work-breakdown · split the work"]
W -.-> C["coding · implement and test"]
W -. "Multiple issues" .-> S["stack-pr · coordinate parallel work"]
S --> C
C --> R["code-review · review the change"]
R --> L["slop-shaking · simplify"]
R -. "Closer review" .-> D["deep-module · check the interface"]
D -.-> T["test-review · check test value"]
D --> L
T --> L
L --> P["github-pr · commit, push, and open a PR"]
When I want to understand a change, I use pr-tour to follow the real diff or
explain-diff for public-interface changes and usage examples, optionally as a visual lesson. eli5 explains unfamiliar topics, and
cram turns study material into a quiz.
npx @solaqua/skul@latest add --global sjquant/ghosts --allOr install the published @solaqua/ghosts package:
npx @solaqua/skul@latest add --global npm:@solaqua/ghosts --allRequires Node.js >=20 and git. --all installs every bundle; --global installs skills at the user level. Skul targets all supported agents by default. Use --agent <name> to choose an agent, or repeat it to choose several.
Most skills maintained here use the ghosts: prefix. For example, invoke
$ghosts:code-review in Codex or /ghosts:code-review in Claude Code.
For installation selectors, use the source path, such as --include skills/code-review.
| Skill | What it helps with |
|---|---|
| coding | Keep implementation simple and tests focused on meaningful behavior. |
| raze | Implement, review and fix, then simplify the change in one workflow. |
| code-review | Review correctness, downstream contracts, design, and complexity; verify material safety assumptions without editing the change. |
| test-review | Find missing test protection and tests that cost more than they protect. |
| docstring-review | Find function summaries that obscure their domain purpose with implementation details. |
| deep-module | Check whether a module hides complexity behind a simple interface. |
| verify-journeys | Verify real user journeys and report observable outcomes, evidence, and coverage gaps. |
| benchmark-checklist | Check performance claims for equivalent work, experimental control, and repeatability. |
| slop-shaking | Remove unnecessary code, content, structure, and UI elements. |
| grill-me | Stress-test a plan or idea through focused questions. |
| work-breakdown | Break a goal into reviewable work packages with dependencies and execution order. |
| task-handoff | Prepare concise handoffs for independent implementation tasks. |
| delegate | Hand work to another Orca agent or supervise it through completion with a concise brief. |
| github-issue | Create or update GitHub issues, sub-issues, and dependencies. |
| github-pr | Commit and push changes, then create or update their pull request. |
| stack-pr | Implement GitHub issues in parallel and build stacked PRs in dependency order. |
| explain-diff | Explain public-interface changes read-only with before/after usage examples, compatibility impact, and optional visual HTML lessons. |
| write-adr | Record an engineering decision, its context, and its rationale. |
| write-glossary | Define project terms and establish a shared vocabulary in GLOSSARY.md. |
| Skill | What it helps with |
|---|---|
| harvest | Extract reusable session and GitHub review experience into Obsidian candidates. |
| curate | Verify candidates and propose Markdown knowledge and INDEX changes. |
| echo | Check the experience index, open only matching guidance, and apply it while working. |
The experience skills use ghosts:harvest, ghosts:curate, and ghosts:echo as their names. See
setup and usage for configuration, source support,
and tests. Echo's library holds curated judgments for selective application.
| Skill | What it helps with |
|---|---|
| ui-design | Explore visual references and image mockups, refine a direction, and deliver design documentation or UI implementation. |
| value-first | Review websites, UX, and READMEs for clear value, brand appeal, trust, and low cognitive load. |
| screen-comments | Document changed PR screens with real screenshots and user-flow comments in the user's language on UI files. |
| Skill | What it helps with |
|---|---|
| deep-research | Investigate a question, test competing explanations, and support conclusions with sources. |
| cold-reader | Write engineering documents that make sense without the preceding conversation. |
| rewrite | Shorten and clarify a prompt while preserving every instruction that affects the result. |
| make-skill | Create or revise a skill with only the instructions needed for its intended behavior. |
| eli5 | Explain a topic to a beginner through a self-contained visual HTML lesson. |
| humanize-korean | Smooth mechanical Korean phrasing while preserving meaning, facts, and tone. |
| fluent-korean | Write Korean responses and documents with clear actions, conditions, and outcomes. |
| Skill | What it helps with |
|---|---|
| vibe-notion | Manage Notion pages, databases, blocks, users, and comments through the CLI. |
These are installed from pinned upstream sources in the registry and keep their upstream names, without ghosts:.
| Skill | What it helps with |
|---|---|
| context7-cli | Fetch current library documentation and manage Context7 skills and configuration. |
| insane-search | Retrieve public web content through alternative access methods when ordinary fetching fails. |
| obsidian-cli | Search and manage Obsidian notes, tasks, properties, and plugins. |
| sentry-cli | Inspect Sentry issues, events, projects, and organizations through the CLI. |
| cram | Turn study material into flashcards and quizzes in a standalone HTML file. |
| pr-tour | Create an HTML reading guide that follows a PR's code flow and real diff. |
| Local component | Upstream | Pinned revision | License |
|---|---|---|---|
writing/skills/humanize-korean/references |
epoko77-ai/im-not-ai, skills/humanize-korean/references |
9747f03; selected rule and comma-metric fixes from b37e020 |
MIT |
tools/skills/vibe-notion |
devxoul/vibe-notion, skills/vibe-notion |
89c6643 |
MIT, declared in the upstream README |