Continue.dev is an open-source IDE assistant (VS Code, JetBrains). It supports custom commands, context providers, and prompt libraries — open-forge plugs into all three.
-
Clone open-forge:
git clone https://github.com/zhangqi444/open-forge ~/code/open-forge -
Generate the Continue-flavored config:
cd ~/code/open-forge ./scripts/build-dist.sh continue # Outputs: # dist/continue/config.snippet.yaml (paste into your continue config) # dist/continue/prompts/*.md (per-recipe prompt files)
-
Merge
dist/continue/config.snippet.yamlinto your~/.continue/config.yaml:# ~/.continue/config.yaml contextProviders: - name: file params: baseDir: ~/code/open-forge/plugins/open-forge/skills/open-forge prompts: - name: self-host description: "Deploy a self-hostable app via open-forge recipes" systemMessage: | You are an expert at deploying self-hostable open-source apps. Follow the open-forge skill at ~/code/open-forge/plugins/open-forge/skills/open-forge/SKILL.md. Use the recipe under references/projects/ that matches the user's request; if no recipe exists, follow the Tier 2 (live-derived) fallback per CLAUDE.md. slashCommands: - name: deploy description: "Self-host an open-source app" prompt: "self-host"
-
Reload Continue (Cmd+Shift+P → Developer: Reload Window).
Use the slash command in Continue's chat:
/deploy Self-host OpenClaw on AWS Lightsail with Bedrock pre-wired.
Or invoke the prompt by name:
@self-host Self-host Vaultwarden on a Hetzner CX22.
You can also reference specific recipes via @file:
@file references/projects/openclaw.md @file references/infra/aws/lightsail.md Now follow the recipe.
| open-forge concept | Continue equivalent |
|---|---|
AskUserQuestion (structured choice) |
Continue's chat doesn't have first-class structured prompts; agent asks in prose with options listed |
WebFetch |
Continue's @URL context provider — paste a URL, agent reads the content |
mcp__github__issue_write |
Continue's terminal context provider can run gh issue create; or use the prefilled-URL fallback |
| Persistent state file | Filesystem operations work via Continue's tool-use (when enabled in config) |
- Tool-use (agent mode) varies by model. Continue's tool-use is model-dependent; Claude / GPT-4 work well, smaller models may struggle with the multi-step phased workflow.
- No auto-trigger on description match. Continue uses explicit slash commands or
@mentions; the open-forge skill description is informational, not a trigger. - Context provider scoping. The
fileprovider above loads the entirereferences/tree; for large catalogs consider per-recipe context providers (@file references/projects/<name>.md) to keep context tight.
User: /deploy Self-host Vaultwarden on a Hetzner CX22.
Continue: Loading the recipe at references/projects/vaultwarden.md and
infra adapter references/infra/hetzner/cloud-cx.md from the
file context provider.
Combo: Hetzner Cloud CX × Docker.
Phases: preflight → provision → dns → tls → smtp → hardening
→ feedback.
For preflight, I need your Hetzner API token. Three safer
paths (per references/modules/credentials.md):
1. File path (e.g. ~/.secrets/hcloud)
2. Env var name (e.g. HCLOUD_TOKEN)
3. Paste — last resort, will need to rotate after deploy
User: 2, HCLOUD_TOKEN
Continue: [reads $HCLOUD_TOKEN via terminal tool]
[runs hcloud server create ...]
...
Continue's terminal context provider can run gh issue create for posting feedback. Or use the prefilled-URL fallback per references/modules/feedback.md § Submission paths.