forge-initlisted
Install: claude install-skill yerros/context-forge
# forge-init
Set up the Context Forge context system in the user's project: the `context/` folder
with six markdown files plus an agent entry point (`CLAUDE.md` or `AGENTS.md`).
The blank templates are bundled at `${CLAUDE_PLUGIN_ROOT}/skills/forge-init/templates/`.
The guiding principle of this methodology: **the user is the architect, the AI is
the implementation engine.** These files capture the architectural thinking so the
agent stays consistent across sessions and never guesses.
## Argument
Text after the command seeds the setup (e.g. `/forge-init a Next.js SaaS for
invoicing` gives greenfield answers a head start; `/forge-init adopt` skips straight
to reconcile). It never overrides the detector's verdict — run Step 0 regardless.
No argument → just run Step 0 and proceed.
## Step 0: Read the state — ALWAYS run this first
Never assume the project is empty. Run the deterministic state detector from the project
root and branch on its `verdict`. This is read-only and never writes anything:
```bash
bash "${CLAUDE_PLUGIN_ROOT}/skills/forge-init/scripts/detect.sh"
```
The report gives a `verdict` plus the facts behind it (which of the six files exist, how
many unfilled `[placeholder]` markers remain, entry point, decisions/specs presence,
whether a codebase was detected). Act on the verdict — do not re-derive these facts by
hand:
- **`SETUP`** — no `context/` and no files. Fresh install → go to the Mode + Profile step.
- **`ADOPT`** — all six files present and substan