Skill Factory
Skill Factory is an open-source, local-first generator, validator, evaluator, package manager, and lifecycle system for Agent Skills conforming to the Agent Skills open specification.
It bridges the gap between conversational AI prompts and production-grade tools across 8 major AI coding assistants—including Claude Code, Google Antigravity, Cursor, OpenAI Codex, Windsurf, Continue, Cline, and GitHub Copilot.
┌───────────────────────────────────────────────────────────────┐
│ SKILL FACTORY │
│ │
│ [1] Create a new skill (interactive interview) │
│ [2] Search local skill library │
│ [3] Validate an existing skill folder │
│ [4] Run skill automated tests │
│ [5] Evaluate skill quality score (0–100) │
│ [6] Scan skill for secrets and safety issues │
│ [7] Platform health doctor (check AI agents) │
│ [8] Install a skill into an AI tool │
│ [9] List installed skills │
│ [10] Beginner guided walkthrough (step-by-step wizard) │
│ [11] AI-assisted skill creation (natural language) │
│ [12] Curated domain skill packs (browse & install) │
│ [0] Exit │
│ │
│ 100% Python Standard Library • Zero Paid APIs • Local-First │
└───────────────────────────────────────────────────────────────┘
What Skill Factory Does & Does Not Do
To ensure complete transparency and prevent adoption friction, here are the explicit design boundaries of Skill Factory:
What Skill Factory Does
- ✅ Scaffolds Specification-Compliant Skills: Creates standardized folders with validated YAML frontmatter (
name,description), structured markdown instructions (SKILL.md), human documentation (README.md), unit tests (tests/), scripts (scripts/), and cryptographic SHA-256 manifests (.skillfactory/manifest.json). - ✅ Evaluates & Scores Quality (0–100): Grades skills objectively against 5 dimensions: specification compliance, portability, progressive disclosure, test execution, and code safety.
- ✅ Scans for Security Risks & Injections: Static AST scanner flags dangerous calls (
eval,exec,os.system, shell injection). Secrets scanner flags leaked AWS keys, GitHub tokens, and OpenAI/Anthropic keys. Hardened with 7 categories of prompt injection detection and symlink escape defense. - ✅ Manages Complete Lifecycles: Installs skills into official AI agent directories, creates automatic timestamped backup snapshots before updates, performs single-command rollbacks, and cleanly uninstalls skills while preserving user data.
- ✅ Curated Domain Skill Packs: Ships with 5 curated domain packs (Office Productivity, Finance & ERP, HR & Workforce, Project Management, Developer & Security) containing 14 reference skills scoring Grade A (≥90/100).
- ✅ Optional AI-Assisted Skill Creation: Pluggable, provider-neutral synthesis engine (
skillfactory ai-create) supporting local Ollama or compatible models with pre-execution AST safety quarantine and bounded test self-repair. - ✅ Runs 100% Locally & Offline: Requires zero API subscriptions, zero third-party packages, zero account registration, and zero telemetry.
What Skill Factory Does Not Do
- ❌ It is not an autonomous "no-code" app creator: In Offline Mode (
new,interview), Skill Factory generates robust structural scaffolds, validated frontmatter, and test stubs. Domain business logic must be implemented by the developer or by a connected AI assistant. - ❌ It does not invent AI platform features: It adheres strictly to each platform's documented skill discovery paths and capabilities. It does not force autonomous model invocation on platforms that only support manual slash commands or directory indexing.
- ❌ It does not make external network requests: Skill Factory will never secretly upload your skills, metrics, code, or telemetry to a remote server.
Key Features
- ⚡ Zero Third-Party Dependencies: Pure Python standard-library implementation. Runs immediately on any machine with Python 3.9+ without
pip installfriction or virtual environments. - 🔒 Safe & Isolated Execution: Subprocess test runner enforces an isolated 10-second timeout. Zip archive import strictly validates target paths against Zip-Slip traversal attacks (
../), and installation defends against symlink escapes. - 🎯 Dual Console Entry Points: Both
skillfactoryandskill-factoryCLI commands installed out of the box. - 🌐 8 Supported AI Platforms: Native discovery and installation for Claude Code, Google Antigravity, Cursor, OpenAI Codex, Windsurf, Continue, Cline, and GitHub Copilot with collision detection and alias normalization.
- 📦 Automated Backups & Rollback: Every update creates an immutable timestamped backup snapshot, enabling instant rollback if a new version breaks existing behavior.
- 🤖 Optional AI Engine: Synthesizes skills from natural language descriptions via local Ollama or OpenAI-compatible endpoints with AST safety interception and up to 3 automated test self-repair retries.
Supported AI Platforms & Verification Levels
Skill Factory maps to official agent discovery paths defined in skillfactory/data/platforms.json:
| Platform | Project Scope Folder | Global (User) Scope Folder | Verification Level & Activation Model |
|---|---|---|---|
| Claude Code | .claude/skills/<name>/ |
~/.claude/skills/<name>/ |
Verified Autonomous Activation: Host model matches prompt intent to SKILL.md description; also supports manual /<skill-name> command. |
| Google Antigravity | .agents/skills/<name>/ |
~/.gemini/config/skills/ |
Verified Autonomous Activation: Engine injects discovered skills into system prompt <skills> context; model autonomously reads SKILL.md when intent matches. |
| Cursor | .agents/skills/<name>/, .cursor/skills/ |
~/.agents/skills/, ~/.cursor/skills/ |
Verified Directory Discovery: Discovered in workspace or global folder; invoked via manual slash command /<skill-name> or explicit @ context inclusion. |
| OpenAI Codex | .agents/skills/<name>/ |
~/.agents/skills/ |
Simulated CLI Integration: Discovered in .agents/skills/; invoked via CLI syntax $<skill-name>; respects agents/openai.yaml metadata. |
| Windsurf | .codeium/windsurf/skills/<name>/ |
~/.codeium/windsurf/skills/<name>/ |
Verified Directory Discovery: Discovered in Codeium Cascade workspace configuration. |
| Continue | .continue/prompts/<name>/ |
~/.continue/prompts/<name>/ |
Verified Prompt / Tool Folder: Discovered in Continue prompt/custom commands directory. |
| Cline | .cline/skills/<name>/ |
~/.cline/skills/<name>/ |
Verified Workspace Skill Path: Discovered in project .cline/ root directory. |
| GitHub Copilot | .github/skills/<name>/ |
~/.config/github-copilot/skills/<name>/ |
Verified Repository Instruction Path: Discovered in .github/ repository directory. |
Run skillfactory doctor to automatically detect which AI assistants are installed on your workstation.
Quickstart
Prerequisites
Python 3.9 or higher (macOS default Command Line Tools, Linux, or Windows).
1. Launch the Interactive Menu (Beginner Friendly)
# Clone the repository
git clone https://github.com/imnashraf/skill-factory.git
cd skill-factory
# Run the interactive menu (zero flags needed)
python3 -m skillfactory
(On Windows, use py -m skillfactory or python -m skillfactory)
2. Run the Beginner Guided Wizard
# Step-by-step walkthrough checking system health, choosing an assistant, and generating a skill
skillfactory guide
3. Install via pip
# Install from PyPI
pip install agent-skill-factory
# Or install locally from source
pip install .
# Run from anywhere via either command
skillfactory
skill-factory
Detailed operating-system walkthroughs:
- 🍏 macOS Beginner Guide
- 🪟 Windows Beginner Guide
- 🐧 Linux Beginner Guide
- 📖 Comprehensive Beginner Quickstart
- 📚 Beginner Glossary
End-to-End Walkthrough Examples
A. Classic Deterministic Creation & Installation
# Step 1: Create a new skill
skillfactory new --name log-summarizer --description "Analyze server logs and output error distribution summaries."
# Step 2: Validate specification compliance, run tests, and evaluate
skillfactory validate generated-skills/log-summarizer
skillfactory test generated-skills/log-summarizer
skillfactory eval generated-skills/log-summarizer
# Step 3: Install the skill into your AI assistant (e.g. Claude Code or Antigravity)
skillfactory install generated-skills/log-summarizer --platform claude
# Step 4: Verify installed skill status
skillfactory list
B. Curated Domain Skill Packs
# List all 5 available curated skill packs
skillfactory pack list
# Install the entire Finance & ERP Pack into Cursor
skillfactory pack install finance-erp --target cursor
# Verify pack integrity and test coverage
skillfactory pack verify finance-erp
C. Optional AI-Assisted Skill Creation
# Synthesize a skill from natural language with AST safety interception and bounded self-repair
skillfactory ai-create "Compare two attendance logs and identify missing employees, duplicate records, and overtime hours" --name attendance-auditor
Curated Domain Skill Packs (14 Reference Skills)
Skill Factory organizes reference skills into a 3-tier taxonomy (Personal, Professional, Developer) across 5 curated packs. Every included skill is tested and scores Grade A (≥90/100):
1. Office Productivity Pack (office-productivity)
- Excel Comparison Engine (
excel-comparison): Pure standard-library.xlsxparser diffing spreadsheets cell-by-cell. - Smart File Organizer & Renamer (
file-organizer): Categorizes files by MIME/date with dry-run collision defense. - CSV Analytical Report Generator (
csv-to-report): Computes statistical summaries and outputs Markdown, HTML, and JSON. - Markdown Repository Validator (
markdown-doc-validator): Detects broken relative file links, unaligned tables, and untagged fences. - Structured Notes Formatter (
structured-notes-formatter): Transforms raw meeting notes into structured minutes with action item extraction.
2. Finance & ERP Pack (finance-erp)
- Financial Variance Analyzer (
financial-variance-analyzer): Compares budget vs. actual figures, flagging percentage and absolute thresholds. - Invoice Reconciliation (
invoice-reconciliation): Reconciles vendor invoices against purchase orders, identifying mismatched totals and duplicate line items.
3. HR & Workforce Pack (hr-workforce)
- Attendance CSV Validator (
attendance-csv-validator): Audits clock-in/out records for anomalies, missing punches, and negative durations. - Timesheet Reconciliation (
timesheet-reconciliation): Cross-references contractor timesheets against approved project rosters.
4. Project Management Pack (project-management)
- RAID Log Validator (
raid-log-validator): Validates Risk, Assumption, Issue, and Dependency matrices against delivery standards. - Project Status Reporter (
project-status-reporter): Aggregates sprint updates and milestones into executive Markdown summaries.
5. Developer & Security Pack (developer-security)
- Security & Configuration Scanner (
security-config-scanner): AST-based scanner detecting exposed API keys and permissive CORS wildcards. - Dependency Manifest Reviewer (
dependency-manifest-reviewer): Auditsrequirements.txtandpackage.jsonfor unpinned or insecure dependencies. - Test Inventory Analyzer (
test-inventory-analyzer): Inventories automated tests, measuring coverage distribution and skipped test ratios.
CLI Command Reference
All commands are available via skillfactory <command> or skill-factory <command>:
| Command | Description | Example |
|---|---|---|
menu |
Opens the interactive numbered menu (default when run without args). | skillfactory |
guide |
Step-by-step interactive beginner onboarding wizard. | skillfactory guide |
new |
Directly generates a skill with command-line flags. | skillfactory new --name my-tool --desc "Process log files" |
interview |
Guided 5-question terminal interview or non-interactive JSON batch. | skillfactory interview |
ai-create |
Synthesizes a skill from natural language with AST safety interception. | skillfactory ai-create "Extract tables from CSV files" |
pack |
Browse, install, verify, and uninstall curated domain packs. | skillfactory pack list |
validate |
Validates YAML frontmatter, naming, structure, and SHA-256 hashes. | skillfactory validate generated-skills/my-tool |
doctor |
Inspects workstation, Python runtime, and reports installed AI platforms. | skillfactory doctor |
test |
Runs a skill's automated test suite in an isolated subprocess (10s timeout). | skillfactory test generated-skills/my-tool |
eval |
Computes 0–100 quality score and letter grade (Grade A–F). | skillfactory eval generated-skills/my-tool |
scan |
Scans for dangerous AST calls (eval, exec, os.system), prompt injections, and leaked keys. |
skillfactory scan generated-skills/my-tool |
install |
Installs a skill into an AI platform with backup snapshotting. | skillfactory install generated-skills/my-tool --platform cursor |
list |
Lists all installed skills across all detected platforms. | skillfactory list |
update |
Updates an installed skill, detecting local modifications and backing up. | skillfactory update my-tool --platform cursor |
rollback |
Restores an installed skill to its previous backup snapshot. | skillfactory rollback my-tool --platform cursor |
uninstall |
Safely removes an installed skill, preserving user data files. | skillfactory uninstall my-tool --platform cursor |
install-self |
Installs Skill Factory itself into a chosen AI coding assistant. | skillfactory install-self --platform claude |
search |
Sub-millisecond offline keyword/tag search on library. | skillfactory search log |
export |
Packages a skill into a portable .zip with SHA-256 checksum sidecar. |
skillfactory export generated-skills/my-tool |
import |
Safely extracts a .zip skill package with Zip-Slip defense. |
skillfactory import my-tool.zip |
publish |
Generates a standalone repository with README, LICENSE, and CI. | skillfactory publish generated-skills/my-tool --out-dir dist/ |
Security & Privacy Model
Skill Factory adheres to strict host isolation and safety guarantees:
- Local-First & Zero-Telemetry: All operations execute locally. No code, skill names, descriptions, or parameters are ever transmitted over the network.
- AST Danger Detection: Source code is analyzed using Python's abstract syntax tree before execution to intercept forbidden patterns (
eval,exec,os.system,subprocesswithshell=True, and recursive filesystem deletions). - Prompt Injection Defense: Multi-vector screening detects directive overrides, persona hijacks, hidden markup, exfiltration endpoints, and system prompt extraction attacks.
- Symlink Escape Defense: File copying and packaging strictly verify that symlinks do not escape the skill folder boundary to access sensitive host files.
- Secrets & Credential Scanner: Detects exposed AWS keys, GitHub tokens, OpenAI/Anthropic API keys, and PEM certificates.
- Atomic Staging: Files are generated in sibling staging directories (
.tmp-<uuid>) and atomically renamed only upon successful manifest generation, preventing corrupt or half-written skills. - Zip-Slip Defense: Archive extraction strictly verifies resolved destination paths to prevent path traversal attacks (
../). - Subprocess Isolation: Test execution runs in an isolated child process with a strict 10-second timeout, preventing hanging processes or memory exhaustion.
Review our full Security Policy and Security Architecture for vulnerability reporting and threat modeling.
Troubleshooting & FAQ
1. python3: command not found (Windows)
On Windows, the default Python launcher is py. Use:
py -m skillfactory
Or ensure Python is added to your system PATH in the Windows installer.
2. "Destination already exists and is not managed by Skill Factory"
Skill Factory protects your pre-existing files. If a directory was created manually, Skill Factory refuses to overwrite it. To overwrite an unmanaged directory, pass --force:
skillfactory install generated-skills/my-skill --platform claude --force
3. "Skill has local modifications. Use --force to overwrite"
During update, Skill Factory checks current files against the original cryptographic manifest. If you edited files manually, it alerts you to prevent accidental data loss. Passing --force creates a backup snapshot and updates the files.
4. How do I test backward compatibility?
Skill Factory preserves full backward compatibility with the legacy Phase 0 generator:
python3 scripts/create_skill.py --name test-skill --description "Testing backward compatibility"
Governance & Contributing
Skill Factory is an open-source project welcoming community contributions. Please review:
License
MIT License. Copyright (c) 2026 Skill Factory Contributors.
All skills generated by Skill Factory are the 100% exclusive property of the user under whichever license the user selects. Skill Factory does not require viral attribution in your generated skills.
Metadata
Release files for agent-skill-factory 1.2.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| agent_skill_factory-1.2.1.tar.gz | 151.2 kB | With provenance | Details for agent_skill_factory-1.2.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agent_skill_factory-1.2.1-py3-none-any.whl | Python 3 | none | any | Details for agent_skill_factory-1.2.1-py3-none-any.whl |
Total release size: 285.1 kB
Release files / agent_skill_factory-1.2.1.tar.gz
| Download URL | agent_skill_factory-1.2.1.tar.gz |
|---|---|
| Size | 151.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
c6a79f75aaf2d55d03d5d01c23258ee1210aa6fea40c5a432d4b85158b4da667
|
|
BLAKE2b-256 checksum How to use checksums |
5d10981e7c5cd54e74694447bb4dcbcc6e9b2e8ed837a8f505d4ff0ed75d8001
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.12.9
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 20, 2026.
Transparency logRelease files / agent_skill_factory-1.2.1-py3-none-any.whl
| Download URL | agent_skill_factory-1.2.1-py3-none-any.whl |
|---|---|
| Size | 133.9 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
d8d16220b2f4d12d42b574deb816adb13845bc9336d4a923a7fc25cf4b4c7cf3
|
|
BLAKE2b-256 checksum How to use checksums |
5aae31800aae061643c2ea43ff1c30231d033719f9d0ddf49ec25efe76f9dda1
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/6.1.0 CPython/3.12.9
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 20, 2026.
Transparency logSecurity for agent-skill-factory 1.2.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
All 2 files in this release have been uploaded with provenance metadata.
Attestation for agent_skill_factory-1.2.1.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 20, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.