Agent Skill Creator
Turn any workflow into reusable AI agent skills that install on 17 platforms — no spec writing, no coding.
Evidence shows security scanning, validation gates, user confirmation for sharing, rollback via git pull, source attribution via MIT license and author. But least privilege not explicit, user confirmation only partial (when sharing), data flow transparency limited, sensitive data handling not detailed, dependency security not audited, external effects not explicit, rollback only git pull, source attribution only author name. Hence each criterion scored 1.
Self-consistency is good: README, CI, tests align, example skills pass validation. Dependency availability not explicit, failure messages provided (e.g., validation failure, security scan failure). Hence self-consistency 2, dependency availability 1, failure messages 2.
Audience clear (teams, consultants), scenarios diverse (CRM, PR, stock), capability boundaries stated (not buildable list), trigger words described, environment fit (17 platforms). Hence each criterion scored 2.
Information architecture clear (directory structure), install notes detailed, naming rules explicit, examples and FAQ present, known limitations not explicit, license MIT, versioning and changelog present, maintenance responsibility stated (evolve loop). Hence each criterion scored 2, known limitations 1.
Output usability high (generated skills directly usable), marginal value high (cross-platform), cost-benefit reasonable (free open source). Hence each criterion scored 2.
Claim traceability moderate (CI and tests), cross-source corroboration limited (only internal tests), fact-inference separation not explicit. Hence each criterion scored 1.
- Publisher identity unverified, treat with caution.
- Claims of security scanning and validation gates lack implementation details, need further review.
- Dependency security not audited, potential unknown vulnerabilities.
- Data flow transparency limited, sensitive data handling not detailed.
What does this agent do, and when should you use it?
Agent Skill Creator is an open-source tool that converts plain-English workflow descriptions into structured AI agent skills, complete with validation, security scanning, and an eval spec. Generated skills bundle a SKILL.md, functional scripts, eval data, and a cross-platform installer, enabling deployment to 17 platforms including Claude Code, Copilot, Cursor, Windsurf, Codex, and Gemini. The build process follows a five-phase pipeline (Understand, Build, Verify) and enforces four quality gates: validation, security scan, pipeline check, and eval rollout. The tool also provides a skill registry, staleness detection, and team-sharing capabilities. Licensed under MIT.
The tool reads user-provided descriptions or files and produces a complete agent skill package. It parses inputs (text, PDF, URLs, code), runs a five-phase pipeline to generate SKILL.md, executable scripts, an eval spec (run_evals.py), and a cross-platform installer (install.sh). It validates output with validate.py, security_scan.py, and check_pipeline.py. It supports a git-based skill registry (skill_registry.py) for publishing, installing, searching, and staleness reporting. It includes staleness_check.py to check dependency health and schema drift. Generated skills ship with evolve.py for self-maintenance and optional React artifact templates.
- A developer who manually cleans CRM and generates weekly reports can describe the process in natural language and get a reusable skill.
- A team lead wants to standardize deployment checks across the team; they create a skill and share it via git clone.
- An AI consultant builds a skill library for a client, setting up a registry and teaching the team to create and maintain their own skills.
- A data analyst needs to compute technical indicators (RSI, MACD) and buy/sell signals; they can use the bundled stock-analyzer example.
- An engineer wants to convert an existing Python script into a cross-platform skill; they just provide the script path and the tool packages it.
- A security reviewer wants to audit skills for hardcoded secrets or prompt injection; the built-in security scanner automates this.
What are this agent's strengths and limitations?
- Supports 17 platforms including Claude Code, Cursor, Copilot, Gemini, etc., with automatic format adapters.
- Built-in validation, security scanning, and eval rollout ensure skill quality and safety.
- Generated skills ship with an eval spec and evolve.py for self-maintenance and regression testing.
- Skill registry simplifies team sharing and discovery.
- Depends on a runtime AI tool (e.g., Claude Code) to actually use the generated skills; the tool itself is only a generator.
- Requires git and Python 3; Windows support is limited to PowerShell/cmd scripts.
- Security scanner may not catch all threats, so manual review is still advisable for critical skills.
- Eval rollout requires real data, which may involve sensitive information handling.
How do you install or deploy this agent?
macOS/Linux: run curl -fsSL https://raw.githubusercontent.com/FrancyJGLisboa/agent-skill-creator/main/scripts/bootstrap.sh | sh. Windows PowerShell: irm https://raw.githubusercontent.com/FrancyJGLisboa/agent-skill-creator/main/scripts/bootstrap.ps1 | iex. The installer clones the repo to ~/.agents/skills/ and links to detected platforms. Requires git and Python 3.
How do you use this agent?
Open your AI tool and type /agent-skill-creator followed by your workflow description, e.g., /agent-skill-creator Every Friday I clean the CRM export, calculate regional totals, and email a PDF sales report. The tool auto-builds and installs the skill. For examples: clone the repo, then cd references/examples/weekly-crm-report and run python3 scripts/run_pipeline.py --input evals/golden/case-1/input.csv --output /tmp/summary.json, then python3 scripts/run_evals.py --rollout to test.
How does this agent compare with similar options?
Compared to Anthropic's skill-creator, this tool avoids interactive Q&A, handles various input formats, and generates cross-platform installers and eval specs, but skill-creator may be more focused on Claude-specific optimization.