refactor: class components → function components (20 components)

docs/superpowers/specs/2026-08-18-frontier-smol-handoff-design.md

@@ -0,0 +1,74 @@
+# frontier-smol-handoff
+
+Date: 2026-08-18
+Status: draft
+Type: project skill (`.claude/skills/frontier-smol-handoff/SKILL.md`)
+
+## Problem
+
+Agents one-shot implementation, or they write a plan in plan mode and hand that sterile plan to a cheap model. Both waste frontier tokens or starve the cheap model of live exploration context.
+
+The intended workflow: frontier explores, writes a todo list, starts only when confident, is stopped after the first code edit, then a cheap executor continues from the **same session transcript**.
+
+## Decision
+
+Do not invent a custom two-phase protocol. Encode the workflow as a hard-gate skill that uses OMP built-in prewalk and model roles.
+
+| Role | OMP alias | Job |
+|---|---|---|
+| Frontier | the session's active model at invoke (`@default` or `@slow`) | Explore, todo, first directional edit |
+| Cheap executor | `@smol` (`--prewalk-into`, default) | Continue from the live transcript |
+
+Arming (any one is enough):
+
+- `omp --prewalk` (optional `--prewalk-into=@smol`)
+- `prewalk.enabled` in settings
+- agent frontmatter `prewalk: true` or `prewalk: "@smol"`
+- `task.prewalk` / `task.agentPrewalk` for spawned `task` agents
+
+Handoff mechanics are harness-owned: the first workspace-mutating `edit`/`write` (source, not `todo` / notes / `local://`) after a committed todo list switches the session model to the prewalk target. The cheap model inherits the transcript. No plan file.
+
+## Anti-patterns (hard forbid)
+
+- One-shot the whole task on frontier
+- Plan mode or `--plan-yolo` then hand a written plan to `@smol`
+- Spawn cheap `task` agents with a plan and no exploration context
+- Frontier keeps editing after the first mutating write
+- `@smol` starts a new architecture after the switch
+
+Plan mode clears prewalk. If plan mode is on, this skill does not apply.
+
+## Hard gates
+
+1. No workspace-mutating `edit`/`write` until a phased todo list exists (named phases, concrete tasks, not "implement the feature") and exploration is declared complete.
+2. First workspace-mutating tool call is the last frontier action. Keep it small and directional.
+3. After the switch, `@smol` executes the existing todo. No new research, no new design.
+4. If prewalk is not armed when the skill is invoked: tell the user to enable it, or stop after the first edit and ask them to switch to `@smol`. Do not keep going on frontier.
+
+## Trigger
+
+Explicit only. User invokes `/skill:frontier-smol-handoff` or names the workflow. Not auto-loaded.
+
+## Skill shape
+
+Single file: `.claude/skills/frontier-smol-handoff/SKILL.md`.
+
+- Frontmatter `name` + `description` (when-to-use only; no workflow summary)
+- Role table
+- Linear steps + the four hard gates
+- Rationalization table and red flags (discipline skill)
+- Contrast with `--plan-yolo`
+- No extra agent definition, no project `prewalk.enabled` flip unless the user asks
+
+## Testing (writing-skills TDD)
+
+Discipline skill. RED: pressure scenarios without the skill. GREEN: same scenarios with the skill. REFACTOR: close loopholes from observed rationalizations.
+
+Expected baseline failures: one-shot under time pressure; plan-then-smol because it looks cheaper; keep editing on frontier after the first change; hand a written plan to a `task` spawn.
+
+## Non-goals
+
+- Changing OMP defaults for this repo
+- A custom project agent
+- Auto-triggering on every multi-step task
+- Replacing `writing-plans` / `executing-plans` for users who want those

package.json

@@ -135,8 +135,6 @@
   ],
   "homepage": "https://github.com/makebrainwaves/BrainWaves/",
   "devDependencies": {
-    "@babel/plugin-proposal-class-properties": "^7.10.4",
-    "@babel/plugin-proposal-decorators": "^7.10.5",
     "@testing-library/dom": "^10.4.1",
     "@testing-library/jest-dom": "^6.4.0",
     "@testing-library/react": "^16.0.0",