A vendor-neutral, privacy-safe skill suite for forward-deployed engineering: learn a workflow, choose the simplest reliable fix, introduce it safely, and prove its value with real results.
The suite deliberately treats “no AI needed” and “stop” as successful outcomes. It keeps people in control of consequential actions and requires measured results before expansion.
| Skill | Purpose |
|---|---|
fde-copilot |
Start here and route the work through the lifecycle |
fde-discover-workflow |
Observe, measure, and improve the current process |
fde-design-solution |
Choose the simplest reliable intervention |
fde-deploy-solution |
Test, embed, control, and introduce the solution |
fde-review-value |
Measure payback and decide to expand, fix, stop, or wait |
Clone the repository, then copy all five directories under skills/ into your agent’s skills directory. Keep the five directories as siblings because fde-copilot routes between them with relative paths.
Common locations:
- Claude Code:
~/.claude/skills/ - Codex:
~/.codex/skills/ - Other AGENTS.md-compatible tools: use the tool’s configured skills directory, or keep this repository available and point the agent to
skills/fde-copilot/SKILL.md.
Start with either invocation style:
# Claude Code
/fde-copilot Improve one costly workflow in plain language.
# Codex and other $skill hosts
Use $fde-copilot to improve one costly workflow in plain language.
Every skill uses portable SKILL.md frontmatter with only name and description. Product-specific display metadata is isolated in agents/openai.yaml. Repository guidance lives in AGENTS.md, while CLAUDE.md imports the same guidance so the two files cannot drift.
This public repository was created from a sanitized export, not by publishing the source repository or its history. Examples are generic, and the initial commit uses non-personal maintainer metadata. Do not add customer names, organization names, personal paths, email addresses, phone numbers, credentials, production URLs, or private datasets.
MIT