A collection of Cursor Agent Skills shared for the tech community.
Skills are reusable instruction sets that teach AI agents how to perform specialized workflows — code review, documentation, automation, and more.
Skills live under common-skills/. Each skill is a directory with a required SKILL.md file.
common-skills/
├── ask-oracle/
│ └── SKILL.md
├── article-polish/
│ └── SKILL.md
├── article-workflow/
│ ├── README.md
│ ├── brief/
│ │ └── SKILL.md
│ ├── clean-sources/
│ │ └── SKILL.md
│ ├── section-review/
│ │ └── SKILL.md
│ ├── evidence-pool/
│ │ └── SKILL.md
│ ├── ai-edit-pass/
│ │ └── SKILL.md
│ ├── global-review/
│ │ └── SKILL.md
│ ├── main-draft/
│ │ └── SKILL.md
│ ├── visual-plan/
│ │ └── SKILL.md
│ ├── style-bible/
│ │ └── SKILL.md
│ ├── polish/
│ │ └── SKILL.md
│ ├── final-review/
│ │ └── SKILL.md
│ ├── publish/
│ │ └── SKILL.md
│ └── skill-maker/
│ └── SKILL.md
├── distill/
│ └── SKILL.md
├── mermaid-lint/
│ ├── SKILL.md
│ ├── validate-mermaid.py # Extracts mermaid blocks, drives the worker
│ └── mermaid-worker.mjs # Renders every block in one browser session
├── open-source-contribution/
│ └── SKILL.md
└── tech-doc-html/
├── SKILL.md
├── references/ # Design system, component templates, security rules
├── scripts/ # Validation helpers
└── assets/ # Example output
See common-skills/README.md for authoring guidelines.
- Clone this repository or copy the skill directory you need.
- Place skills in one of these locations:
- Personal —
~/.cursor/skills/<skill-name>/(available across all projects) - Project —
.cursor/skills/<skill-name>/(shared with the repository)
- Personal —
- Cursor discovers skills automatically from the
SKILL.mdfrontmatter.
For the article-workflow group, copy the entire directory:
~/.cursor/skills/article-workflow/
Skills are plain markdown. You can adapt the instructions for other AI coding tools that support custom system prompts or skill files.
| Skill | Description |
|---|---|
| ask-oracle | Produces a concise brief containing the original request and all decision-relevant context, reserving technical judgment for an expert oracle. |
| article-polish | Article polishing with quick / normal / refined modes. Derivative work based on baoyu-translate. |
| article-workflow | A phased article optimization workflow with 13 skills — from brief generation through final publication. See workflow README for phase order and usage. |
| distill | Reviews agent sessions for evidence-backed harness improvements, pruning stale or conflicting rules before strengthening specs, tools, checks, skills, or CI. |
| mermaid-lint | Validates and fixes mermaid diagrams in markdown. Renders every block against the real mermaid renderer and reports all failures in one pass. Original skill design. |
| open-source-contribution | Open-source contribution hygiene: privacy scanning, Git history cleanup, installer hardening, autoreview, and safe push/PR validation. |
| tech-doc-html | Interactive single-file HTML from technical design docs. Original skill design; visual style inspired by html-effectiveness. |
A manually invoked skill that frames the user's request and all known decision-changing context, then asks an expert oracle to supply the technical judgment.
Polishes and improves writing with three modes (quick / normal / refined). Supports style presets, audience tuning, long-document chunking, and persistent preferences via EXTEND.md.
A phased article optimization workflow. Each skill handles one phase — from brief generation through final publication. Skills are designed to be used in sequence, but each can also be invoked independently. See the workflow README for details.
Recommended order:
| Phase | Skill | Purpose |
|---|---|---|
| 0 | article-workflow-brief |
Generate an editorial brief |
| 0.5 | article-workflow-clean-sources |
Clean oral draft transcription errors |
| 1 | article-workflow-section-review |
Section-by-section narrative review |
| 2 | article-workflow-evidence-pool |
Fact-checking and material pool |
| 3 | (Author self-read and direct editing) | Author directly edits .article-workflow/00-cleaned-sources/ |
| 3.5 | article-workflow-ai-edit-pass |
AI-assisted editing based on confirmed decisions |
| 4 | article-workflow-global-review |
Whole-article coherence review |
| 4.5 | article-workflow-main-draft |
Integrate sections into a continuous draft |
| 4.6 | article-workflow-visual-plan |
Illustration and visual aid planning |
| 5.0 | article-workflow-style-bible |
Extract a style bible |
| 5 | article-workflow-polish |
Multi-round polishing |
| 6 | article-workflow-final-review |
Final review and reader testing |
| 8 | article-workflow-publish |
Sync to publishing channels |
The article-workflow-skill-maker is a meta skill for turning a manually executed phase into a reusable workflow skill.
Replays the current session as a Harness Engineering retrospective, including failures, successful shortcuts, and interaction friction. Before adding guidance, it audits existing rules for staleness, duplication, conflicts, deterministic replacements, and misplaced scope. It then proposes at most five changes across specs, context, skills, tools, environment, tests, lint, pre-commit, CI, and evals behind one compact approval gate.
Finds every mermaid diagram in one or more markdown files, validates it, and fixes the
broken ones. Unlike the other skills here it ships executable helpers, so it needs
Node.js and @mermaid-js/mermaid-cli
available on PATH; the skill will not install them for you.
Two design decisions are worth calling out, because the obvious alternatives are worse:
- It renders each diagram instead of only parsing it.
mermaid.parse()covers the parse phase only, so errors raised while rendering slip through — an invalid gantt date such asnotadateparses fine but fails to render. Rendering answers the question a document author actually has: will this diagram show up? - It renders the whole batch in a single browser session. Spawning one Chromium per
diagram costs roughly 1.7s each; sharing a session brings the marginal cost down to
about 12ms, so 60 diagrams take ~2s instead of ~100s. Running
mmdcover the markdown file directly would also share a session, but it aborts on the first bad diagram, which defeats the point of a linter.
Block extraction follows CommonMark fence rules. A deliberately broken example nested inside a longer fence is not reported as a real error, and directive-style blocks, fences carrying an info string, and tilde fences are all recognized.
Standardizes open-source contribution cleanup and release checks for coding agents: scan file content and Git metadata, remove local paths and private identities, harden installers, preserve streaming behavior in local proxies, run gitleaks/pre-commit/tests, use autoreview as a closeout gate, and verify history rewrites before pushing.
An original Cursor skill that converts technical specs into interactive HTML visualizations. The agent picks components per section (Mermaid diagrams, SVG sliders, comparison tables, risk matrices), runs Mermaid security checks and Playwright QA, and applies a visual style inspired by html-effectiveness.
Triggers: generate technical design HTML, architecture diagram pages, interactive RFC visualizations.
Includes: design system reference, component pattern library, example output, security validation scripts.
This is a personal skills collection. Feel free to fork, adapt, and use the skills under the MIT License.
If you find a bug or have a suggestion, open an issue or pull request.
This repository is released under the MIT License.
Derivative work based on baoyu-translate from baoyu-skills (MIT, Copyright Jim Liu). Repurposes the workflow for writing improvement. Independent project.
Original skill designs for a phased article optimization workflow. Each skill covers one phase — from brief generation through publication. The visual planning phase references a generic article-illustrator skill for prompt construction rules.
Original skill design. Drives mermaid through
@mermaid-js/mermaid-cli (MIT) at runtime;
neither project's code is vendored here.
Original Cursor skill design. Visual style inspired by html-effectiveness (Apache-2.0, Copyright Anthropic PBC). Style patterns used in references/design_system.md and references/component_patterns.md. Full Apache-2.0 text.