Turn any workflow into
reusable AI agent software.
Describe a workflow in plain English — or hand over a PDF, a link, a script. Get back a complete, validated, security-scanned agent skill with functional code, its own eval spec, and a cross-platform installer. No spec writing. No prompt engineering. No coding.
Genuine output from a real run — validate → security_scan → eval --rollout on a generated skill.
Five phases, three stages, two hard gates.
You don't need to understand any of this to use it — the agent reads your material, uncovers the implicit requirements, and builds from a deeper spec than what you typed.
Discovery + Design
Reads all your material, researches APIs, and derives what you actually need — not just what you typed.
- Phase 1 Discovery — compare APIs by cost, rate limits, docs; decide with justification
- Phase 2 Design — 4–6 priority use cases, artifact assessment, eval criteria (3–6 checks + 3 golden cases)
Architecture + Detection
Decides simple skill vs. multi-agent suite, then writes activation keywords so agents actually discover it.
- Phase 3 Architecture — single SKILL.md or complex component suite
- Phase 4 Detection — ≤1024-char description with domain keywords for cross-platform activation
Implementation + Gates
Generates every file, then blocks delivery unless it passes validation and a security scan.
- Phase 5 SKILL.md, AGENTS.md, scripts, references, eval spec, install.sh, README
- Gates spec validation, security scan (keys, injection), pipeline check, eval
--validate
Three runnable examples
Each ships in references/examples/ and passes the same gates the creator applies to skills it generates — validate.py, check_pipeline.py, and its own bundled eval spec.
weekly-crm-report
"Clean this CRM export and total sales by region."
references/examples/weekly-crm-report/
pr-blocker-summarizer
"Summarize my open PRs, blockers first."
references/examples/pr-blocker-summarizer/
stock-analyzer
"Analyze AAPL with RSI and MACD."
references/examples/stock-analyzer/
Built-in gates that others leave to you
| Agent Skill Creator | Hand-writing a SKILL.md | Anthropic's skill-creator | |
|---|---|---|---|
| Time to a working skill | ~minutes, one prompt | Hours of spec + iteration | Minutes (interactive Q&A) |
| Coding required | None — it writes the code | Yes | Some |
| Cross-platform install | 17 platforms, auto-detected | One tool, by hand | Claude-focused |
| Validation + security scan | Yes, hard gates | Manual | Partial |
| Ships an eval spec | Yes, per skill | No | No |
| Optimizable (autoresearch handoff) | Yes | No | No |
| Input you can hand it | Prose, PDF, URL, code, transcript | You write it from scratch | Guided prompts |
Artifacts, and a metric for every skill
Interactive React artifacts, no React written
When output is visualizable — time series, comparisons, KPIs, tables — Phase 2 inlines one of four bundled templates (line-chart, bar-chart, kpi-cards, data-table) plus the artifact protocol into the generated SKILL.md. Hosts that don't render artifacts get honest degradation: fenced code, unchanged markdown analysis. Suppress with --no-artifact, or force one with --artifact <name>.
Every skill ships its own metric
Each generated skill carries evals/<name>.eval.md plus scripts/run_evals.py — 3–6 binary checks and ≥3 golden cases, seeded from your own files. run_evals.py exits non-zero on failure and drops into CI. --rollout --promote captures the first passing baseline. On by default; skip with --no-eval.
Install, then describe what you do
No clone, no pip, no API key to get started — just git and any one of 17 supported tools.
Then open your AI tool and describe what you do:
Auto-detected on install: