Generate and upgrade Claude Code & Antigravity SKILL.md files — with live best practices sync, 10-marketplace discovery, built-in skill upgrader, and self-validation — in minutes.
Built by Lijin Nair · @lijinnair
# Clone into your Antigravity skills directory
git clone https://github.com/lijinnair/claude-code-skillforge ~/.gemini/antigravity/skills/claude-code-skillforge
# Or into native Claude Code
git clone https://github.com/lijinnair/claude-code-skillforge ~/.claude/skills/claude-code-skillforgeThen say: "Build a new skill", "Upgrade this skill", or "Scan my skills" — or run /claude-code-skillforge
Claude Code Skillforge is a meta-skill that builds, upgrades, and scans Claude Code and Antigravity skills. Give it a raw idea, messy workflow, or an existing skill to upgrade, and it outputs a deploy-ready SKILL.md file that follows every official best practice.
Three modes — all start with a live best practices sync, then branch:
Build Mode — "Build a new skill":
| Step | What happens |
|---|---|
| 1 | Live Best Practices Sync — Fetches latest rules from code.claude.com and antigravity.google |
| 2 | Intake — Collects your skill name, category, ecosystem, triggers, inputs, and outputs |
| 3 | Discovery — Searches 10 marketplaces in parallel to find existing similar skills |
| 4 | Front Matter — Engineers a spec-compliant, sub-1024-char metadata block |
| 5 | SOP Translation — Converts your workflow into an imperative execution pipeline |
| 6 | Self-Validation — Audits its own output against a 27-item, 3-section checklist before delivery |
Upgrade Mode — "Upgrade this skill":
| Step | What happens |
|---|---|
| U1 | Ingest — Reads and parses the existing SKILL.md |
| U2 | Diagnostic Audit — Runs 27-item checklist, reports Pass/Warning/Fail per item |
| U3 | Upgrade & Fix — Applies all fixes while preserving original intent |
| U4 | Deliver — Change summary + upgraded SKILL.md |
Scan Mode — "Scan my skills":
| Step | What happens |
|---|---|
| S1 | Discover — Finds all SKILL.md files in your skills directories |
| S2 | Quick Audit — Runs key checks on each skill |
| S3 | Health Report — Outputs a table sorted by most issues first |
flowchart TD
A([User invokes Skillforge]) --> B[Step 0: Update Check]
B --> C[Step 1: Sync Live Best Practices]
C --> D{Step 1.5: Mode Detection}
D -->|"build / create / new"| E[Build Mode]
D -->|"upgrade / audit / fix"| F[Upgrade Mode]
D -->|"scan / health check"| G[Scan Mode]
E --> E1[Intake & Scoping]
E1 --> E2[Discovery — 10 Sources]
E2 --> E3[Front Matter Engineering]
E3 --> E4[SOP Translation]
E4 --> E5[Scaffold Generation]
E5 --> E6[Validate & Deliver]
F --> F1[Ingest Existing SKILL.md]
F1 --> F2[Diagnostic Audit — 27 Items]
F2 --> F3[Upgrade & Fix]
F3 --> F4[Deliver Upgraded Skill]
G --> G1[Discover SKILL.md Files]
G1 --> G2[Quick Audit Each Skill]
G2 --> G3[Health Report Table]
style A fill:#7c3aed,color:#fff
style D fill:#f59e0b,color:#000
style E fill:#3b82f6,color:#fff
style F fill:#10b981,color:#fff
style G fill:#ef4444,color:#fff
flowchart LR
subgraph sync [" Live Sync "]
S1[Fetch 5 docs URLs<br/>in parallel]
S1 --> S2{All succeeded?}
S2 -->|Yes| S3[Present best<br/>practices summary]
S2 -->|No| S4[Fall back to<br/>cached knowledge]
end
subgraph intake [" Intake "]
I1[Collect name,<br/>category, ecosystem]
I1 --> I2[Define triggers,<br/>inputs, outputs]
end
subgraph discovery [" Discovery "]
D1[Search 10 sources<br/>in parallel]
D1 --> D2{Match found?}
D2 -->|Yes| D3[Present report —<br/>Customise or Build new?]
D2 -->|No| D4[Proceed to build]
end
subgraph build [" Build "]
B1[Engineer<br/>front matter]
B1 --> B2[Translate workflow<br/>to SOP steps]
B2 --> B3[Generate scaffold<br/>+ SKILL.md]
end
subgraph validate [" Validate "]
V1[Run 27-item<br/>checklist]
V1 --> V2{All pass?}
V2 -->|No| V3[Fix failures]
V3 --> V1
V2 -->|Yes| V4[Deliver output]
end
sync --> intake --> discovery --> build --> validate
flowchart TD
U0([Existing SKILL.md provided]) --> U1[Parse front matter & body]
U1 --> U2[Run 27-item checklist]
U2 --> U3{Results}
U3 --> P[Pass items]
U3 --> W[Warning items]
U3 --> F[Fail items]
P & W & F --> U4[Present diagnostic report]
U4 --> U5{User confirms<br/>upgrade?}
U5 -->|No| U6([End])
U5 -->|Yes| U7[Apply fixes —<br/>preserve original intent]
U7 --> U8[Change summary +<br/>upgraded SKILL.md]
style U0 fill:#10b981,color:#fff
style P fill:#22c55e,color:#fff
style W fill:#f59e0b,color:#000
style F fill:#ef4444,color:#fff
flowchart LR
S0([User says<br/>'Scan my skills']) --> S1
subgraph discover [" Discover "]
S1[Scan ~/.claude/skills/]
S2[Scan ~/.gemini/antigravity/skills/]
S1 & S2 --> S3[List all<br/>SKILL.md files]
end
subgraph audit [" Audit Each Skill "]
S3 --> A1[Front matter check]
S3 --> A2[Name constraints]
S3 --> A3[Description format]
S3 --> A4[Body line count]
S3 --> A5[Imperative verbs]
end
subgraph report [" Report "]
A1 & A2 & A3 & A4 & A5 --> R1[Health Report Table<br/>sorted by most issues]
end
style S0 fill:#ef4444,color:#fff
flowchart TD
Q([Skill name]) --> P[Parallel search — all 10 sources]
P --> M1[Smithery]
P --> M2[SkillsMP]
P --> M3[SkillsLLM]
P --> M4[SkillHub]
P --> M5[Antigravity Skill Vault]
P --> M6[Awesome Skills]
P --> M7[GitHub Topics]
P --> M8[Composio]
P --> M9[AI Templates]
P --> M10[Awesome Claude Skills]
M1 & M2 & M3 & M4 & M5 & M6 & M7 & M8 & M9 & M10 --> R{Results}
R -->|No match| B[Build from scratch]
R -->|Match found| C[Discovery Report —<br/>Customise or Build new?]
style Q fill:#7c3aed,color:#fff
style P fill:#3b82f6,color:#fff
flowchart TD
V([Generated SKILL.md]) --> C1
subgraph core [" Core Quality — 16 checks "]
C1[Front matter < 1024 chars]
C2[Standard fields only]
C3[Name constraints]
C4[Description format]
C5[Body < 500 lines]
C6[Imperative verbs]
C7[Progressive disclosure]
C8[...]
end
subgraph code [" Code & Scripts — 8 checks "]
C9[Error handling]
C10[No voodoo constants]
C11[Packages listed]
C12[...]
end
subgraph post [" Post-Delivery — 3 checks "]
C13[Test across models]
C14[Evaluation scenarios]
C15[Real usage tests]
end
core & code & post --> D{All pass?}
D -->|No| FIX[Fix failures] --> V
D -->|Yes| SHIP([Deliver to user])
style V fill:#f59e0b,color:#000
style SHIP fill:#22c55e,color:#fff
Most Claude Code and Antigravity skills are written like essays. Claude Code Skillforge enforces Progressive Disclosure — the principle that every token in the context window must earn its place:
- Minimal footprint in the context window
- Deterministic logic offloaded to
scripts/ - Examples and references loaded only when needed
- Strict imperative pipelines, not conversational prose
The result: skills that are faster, more reliable, and cheaper to run.
- Dual ecosystem — Generates skills for Claude Code or Antigravity
- Live best practices sync — Always builds against the latest official spec
- Skill upgrader — Feed it any existing SKILL.md and get a diagnostic audit + automated upgrade to latest best practices
- Skill scanner — Say "Scan my skills" to audit all installed skills at once and get a health report table with issues ranked by severity
- 10-source discovery — Searches Smithery, SkillsMP, SkillsLLM, SkillHub, Composio, AI Templates, GitHub Topics, Awesome Claude Skills, and more before building from scratch
- Graceful fallback — If docs are unreachable, surfaces cached version and asks before proceeding
- Self-validating — 27-item, 3-section checklist (Core quality, Code & scripts, Post-delivery) catches errors before you see the output
- Pattern library — Teaches 8 authoring patterns: degrees of freedom, feedback loops, templates, examples, conditional workflows, checklists, verifiable intermediates, defaults over options
- Evaluation-ready — Ships with 5 evaluation scenarios covering Build, Upgrade, and Scan modes
- Token-optimized — The skill itself practices what it preaches (< 250 lines, ~1,800 tokens)
- Self-updating — Checks for newer versions on GitHub before every run, with a one-line update notice
claude-code-skillforge/
├── SKILL.md ← The core SOP (v5.11.0)
├── VERSION ← Current version string (used by self-update)
├── README.md ← You are here
├── CHANGELOG.md ← Full version history
├── CONTRIBUTING.md ← How to contribute
├── LICENSE ← MIT
├── skill.json ← Marketplace metadata
├── examples/
│ ├── auditing-seo/SKILL.md
│ ├── formatting-commits/SKILL.md
│ ├── sample-upgrade-report.md
│ └── sample-scan-report.md
└── evaluations/
├── eval-simple-skill.json
├── eval-skill-with-scripts.json
├── eval-advanced-skill.json
├── eval-upgrade-mode.json
└── eval-scan-mode.json
Build Mode outputs:
- Auditing SEO — Audits any URL for title tags, meta descriptions, heading structure, and content quality
- Formatting Commits — Generates Conventional Commits messages from staged diffs
Upgrade & Scan Mode outputs:
- Sample Upgrade Report — Diagnostic audit with Pass/Warning/Fail + change summary
- Sample Scan Report — Health report table across all installed skills
See CONTRIBUTING.md. All contributions welcome.
Lijin Nair — AI Systems Builder & Prompt Architect
- GitHub: @lijinnair
- Web: lijinnair.com
If Claude Code Skillforge saved you time, star the repo — it helps others find it.