feat: add flare-edge-cli agent skill

.gitignore

@@ -3,7 +3,7 @@
 .env.*
 .idea/
 .vscode/
-flare-edge-cli
+/flare-edge-cli
 bin/
 coverage.out
 dist/

README.md

@@ -91,6 +91,14 @@ Prebuilt release archives are attached automatically for:
 
 Release downloads are published on the [GitHub Releases](https://github.com/Gladium-AI/flare-edge-cli/releases) page.
 
+Install the Claude Code skill:
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Gladium-AI/flare-edge-cli/main/install-skill.sh | sh
+```
+
+That installs the skill to `~/.claude/skills/flare-edge-cli` by default. Override the target directory with `SKILLS_DIR=/path/to/skills`.
+
 Build from source from the repository root:
 
 ```bash

install-skill.sh

@@ -0,0 +1,69 @@
+#!/bin/sh
+set -eu
+
+REPO="${REPO:-Gladium-AI/flare-edge-cli}"
+REF="${REF:-main}"
+SKILL_NAME="${SKILL_NAME:-flare-edge-cli}"
+SKILL_PATH="${SKILL_PATH:-skills/${SKILL_NAME}}"
+SKILLS_DIR="${SKILLS_DIR:-${CLAUDE_CODE_SKILLS_DIR:-$HOME/.claude/skills}}"
+
+download() {
+	url="$1"
+	out="$2"
+
+	if command -v curl >/dev/null 2>&1; then
+		curl -fsSL "$url" -o "$out"
+		return
+	fi
+
+	if command -v wget >/dev/null 2>&1; then
+		wget -q "$url" -O "$out"
+		return
+	fi
+
+	echo "Error: curl or wget is required" >&2
+	exit 1
+}
+
+copy_skill() {
+	src="$1"
+	dst="${SKILLS_DIR}/${SKILL_NAME}"
+
+	mkdir -p "$SKILLS_DIR"
+	rm -rf "$dst"
+	cp -R "$src" "$dst"
+	printf 'Installed skill to %s\n' "$dst"
+}
+
+if [ -n "${SOURCE_DIR:-}" ]; then
+	src="${SOURCE_DIR%/}/${SKILL_PATH}"
+	if [ ! -d "$src" ]; then
+		echo "Error: skill directory not found at $src" >&2
+		exit 1
+	fi
+	copy_skill "$src"
+	exit 0
+fi
+
+tmpdir="$(mktemp -d)"
+trap 'rm -rf "$tmpdir"' EXIT INT TERM
+
+archive="$tmpdir/repo.tar.gz"
+url="https://codeload.github.com/${REPO}/tar.gz/refs/heads/${REF}"
+
+download "$url" "$archive"
+tar -xzf "$archive" -C "$tmpdir"
+
+root_dir="$(find "$tmpdir" -mindepth 1 -maxdepth 1 -type d | head -n 1)"
+if [ -z "$root_dir" ]; then
+	echo "Error: could not extract repository archive" >&2
+	exit 1
+fi
+
+src="${root_dir}/${SKILL_PATH}"
+if [ ! -d "$src" ]; then
+	echo "Error: skill directory not found in downloaded archive: $src" >&2
+	exit 1
+fi
+
+copy_skill "$src"

skills/flare-edge-cli/SKILL.md

@@ -0,0 +1,100 @@
+---
+name: flare-edge-cli
+description: Use this skill when the task is to scaffold, validate, build, develop, deploy, inspect, operate, or tear down Go-based Cloudflare Workers with flare-edge-cli. Trigger on requests about Cloudflare edge functions, Go/Wasm Workers, Workers AI in Go, or cleanup of ephemeral Cloudflare resources created through flare-edge-cli.
+license: MIT
+metadata:
+  author: Gladium AI
+  version: 1.0.0
+  category: developer-tools
+  tags:
+    - cloudflare
+    - workers
+    - golang
+    - wasm
+    - workers-ai
+    - deployment
+---
+
+# Flare Edge CLI
+
+Use this skill to operate `flare-edge-cli` safely and consistently.
+
+## Use This Skill For
+
+- Creating a new Go-based Cloudflare Worker project
+- Validating Go-for-Wasm compatibility before build or deploy
+- Building and running local dev flows for Workers
+- Deploying or operating KV, D1, R2, routes, secrets, releases, and logs
+- Creating or testing Go-based Workers AI projects
+- Tearing down disposable Workers and attached side effects
+
+## Core Rules
+
+- Prefer `flare-edge-cli` over raw `wrangler` when the task fits the CLI surface.
+- Prefer `--json` output when another agent or program will consume the result.
+- Keep project work scoped to an explicit `--path` when you are not already inside the generated project directory.
+- Do not run `auth logout --all` unless the user explicitly asks to clear Cloudflare auth.
+- For disposable test environments, finish with `flare-edge-cli teardown` so remote side effects are removed.
+- If a task involves real Cloudflare AI usage, note that local dev still uses remote Workers AI and may incur charges.
+
+## Quick Workflow
+
+1. Verify prerequisites.
+2. Initialize or inspect the project.
+3. Run compatibility and build checks.
+4. Use `dev` for local validation.
+5. Use `deploy` for live rollout.
+6. Use service-specific commands for KV, D1, R2, secrets, routes, logs, and releases.
+7. Use `teardown` for cleanup when the environment is temporary.
+
+## Prerequisites
+
+- `flare-edge-cli` available on `PATH`, or use the repo-local binary/build path.
+- Go installed.
+- Wrangler installed.
+- Cloudflare auth already configured.
+
+For the latest install flow and release-binary behavior, see [references/install-and-release.md](references/install-and-release.md).
+
+## Standard Command Path
+
+For a standard Worker:
+
+```bash
+flare-edge-cli doctor --json
+flare-edge-cli project init my-worker --template edge-http
+flare-edge-cli compat check --path ./my-worker --json
+flare-edge-cli build --path ./my-worker --json
+flare-edge-cli dev --path ./my-worker --local
+flare-edge-cli deploy --path ./my-worker --json
+```
+
+For an AI Worker:
+
+```bash
+flare-edge-cli project init my-ai-worker --template ai-chat
+flare-edge-cli build --path ./my-ai-worker --json
+flare-edge-cli dev --path ./my-ai-worker --local
+flare-edge-cli deploy --path ./my-ai-worker --json
+```
+
+Load [references/ai-workers.md](references/ai-workers.md) when the task is about AI templates, current model defaults, or how to test Workers AI locally.
+
+## Cleanup
+
+If you create temporary infrastructure or disposable test projects, tear them down explicitly:
+
+```bash
+flare-edge-cli teardown --path ./my-worker --json
+```
+
+Use `--keep-bindings` only when the user wants to preserve KV, D1, or R2 resources.
+
+## Troubleshooting
+
+- Start with `flare-edge-cli doctor`.
+- If deployment succeeds but the Worker fails at runtime, use `flare-edge-cli logs tail`.
+- If a command mutates Cloudflare resources, verify whether the target project path and Worker name are correct before rerunning it.
+- If a project mixes manual Wrangler edits with CLI-managed config, inspect both `flare-edge.json` and `wrangler.jsonc`.
+
+For common command sequences and operational guidance, read [references/workflows.md](references/workflows.md).

skills/flare-edge-cli/agents/openai.yaml

@@ -0,0 +1,4 @@
+interface:
+  display_name: "Flare Edge CLI"
+  short_description: "Operate Go-based Cloudflare Workers and AI Workers with flare-edge-cli"
+  default_prompt: "Use flare-edge-cli to scaffold, validate, build, deploy, test, or tear down a Go-based Cloudflare Worker and return the concrete commands, outputs, and cleanup actions."

skills/flare-edge-cli/references/ai-workers.md

@@ -0,0 +1,50 @@
+# AI Workers With Flare Edge CLI
+
+`flare-edge-cli` supports first-class Workers AI scaffolds for Go/Wasm Workers.
+
+## Supported AI Templates
+
+- `ai-text`
+- `ai-chat`
+- `ai-vision`
+- `ai-stt`
+- `ai-tts`
+- `ai-image`
+- `ai-embeddings`
+
+## Current Default Models
+
+- `ai-text` / `ai-chat`: `@cf/moonshotai/kimi-k2.5`
+- `ai-vision`: `@cf/moonshotai/kimi-k2.5`
+- `ai-stt`: `@cf/deepgram/nova-3`
+- `ai-tts`: `@cf/deepgram/aura-2-en`
+- `ai-image`: `@cf/black-forest-labs/flux-2-klein-9b`
+- `ai-embeddings`: `@cf/qwen/qwen3-embedding-0.6b`
+
+## Local Testing Guidance
+
+- `flare-edge-cli dev --local` still uses the remote Cloudflare AI binding.
+- Expect real account usage and potential charges during local testing.
+- First request in a fresh `wrangler dev` session may prompt for Cloudflare account selection.
+
+## Typical AI Workflow
+
+```bash
+flare-edge-cli project init demo-ai --template ai-chat
+flare-edge-cli build --path ./demo-ai --json
+flare-edge-cli dev --path ./demo-ai --local
+```
+
+Then test the local endpoint with a simple request, for example:
+
+```bash
+curl 'http://127.0.0.1:8787/?prompt=Reply%20with%20OK'
+```
+
+## Cleanup
+
+If the AI Worker was created for a temporary run, prefer:
+
+```bash
+flare-edge-cli teardown --path ./demo-ai --json
+```

skills/flare-edge-cli/references/install-and-release.md

@@ -0,0 +1,29 @@
+# Install And Release
+
+## Install Options
+
+One-line installer:
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/Gladium-AI/flare-edge-cli/main/install.sh | sh
+```
+
+Source build:
+
+```bash
+make build
+make install
+```
+
+## Release Assets
+
+GitHub releases publish binaries for:
+
+- `linux/amd64`
+- `linux/arm64`
+- `darwin/amd64`
+- `darwin/arm64`
+- `windows/amd64`
+- `windows/arm64`
+
+Release assets are attached automatically by the GitHub Actions release workflow when a GitHub release is published.

skills/flare-edge-cli/references/workflows.md

@@ -0,0 +1,60 @@
+# Flare Edge CLI Workflows
+
+Use these patterns when operating `flare-edge-cli` projects.
+
+## 1. Bootstrap a Standard Worker
+
+```bash
+flare-edge-cli doctor --json
+flare-edge-cli project init hello-edge --template edge-http
+flare-edge-cli compat check --path ./hello-edge --json
+flare-edge-cli build --path ./hello-edge --json
+flare-edge-cli dev --path ./hello-edge --local
+flare-edge-cli deploy --path ./hello-edge --json
+```
+
+## 2. Bootstrap an AI Worker
+
+```bash
+flare-edge-cli project init hello-ai --template ai-chat
+flare-edge-cli build --path ./hello-ai --json
+flare-edge-cli dev --path ./hello-ai --local
+flare-edge-cli deploy --path ./hello-ai --json
+```
+
+## 3. Operate an Existing Project
+
+```bash
+flare-edge-cli project info --path ./hello-edge --json
+flare-edge-cli compat check --path ./hello-edge --json
+flare-edge-cli build --path ./hello-edge --json
+flare-edge-cli logs tail --path ./hello-edge
+```
+
+## 4. Provision Data Resources
+
+```bash
+flare-edge-cli kv namespace create CACHE --path ./hello-edge --json
+flare-edge-cli d1 create DB --path ./hello-edge --json
+flare-edge-cli r2 bucket create FILES --path ./hello-edge --json
+```
+
+## 5. Safe Cleanup
+
+For temporary resources created during tests or agent workflows:
+
+```bash
+flare-edge-cli teardown --path ./hello-edge --json
+```
+
+If the local project should also be removed:
+
+```bash
+flare-edge-cli teardown --path ./hello-edge --delete-project --json
+```
+
+## Notes
+
+- Prefer `--json` for machine consumption.
+- Prefer `--path` over relying on the current working directory when the target project is not obvious.
+- Use `doctor` early if auth or tooling health is uncertain.