feat(template): disable adapter-node precompress to shrink tarball (#125)

## Summary

- Flip `adapter()` → `adapter({ precompress: false })` in
`template/svelte.config.js`.
- Drops 84 precompressed siblings (`*.br` + `*.gz`) from both shipped
images.
- Each `dist*/` artifact: **2.0M → 1.5M** (–25%, ~520K saving per image,
~1.04 MB total).
- Closes #120.

## Why

Per issue #120, the precompressed siblings cost ~520K per image but buy
negligible runtime CPU savings for the canonical local-first deployment
(`npx @forgeplan/web start` on `127.0.0.1`, single user, no CDN). After
PR #117 minified the server bundle, precompressed client assets were the
largest fixed cost left in the experimental tarball.

Decision (PRD-031 + EVID-036): chose **Option A** from #120 (disable at
adapter level) over B (build.mjs strip — strictly more code for same
runtime behaviour) and C (keep `.br`, drop `.gz` — half the win for
~zero additional safety).

## Trade-off (R-1, documented in PRD-031)

Sirv serves assets uncompressed when no precompressed sibling exists.
Verified via direct `Accept-Encoding: br, gzip` probe against a live
`npx ... start` server in a scratch dir:

```
HTTP/1.1 200 OK
Content-Length: 3047
```

(Same response with and without `Accept-Encoding`. No `Content-Encoding`
header in either case.) For loopback `127.0.0.1` deployments the cost is
negligible. Mitigation paths if a real complaint surfaces: revert this
flag, or wire on-the-fly compression in `template/src/hooks.server.ts`.

## Test plan

- [x] `npm run build` — both images emit (`dist/ ready (1.40M)`,
`dist-nightly/ ready (1.40M)`).
- [x] `find dist dist-nightly -name '*.br' -o -name '*.gz' | wc -l` → 0
(was 84).
- [x] `du -sh dist dist-nightly` → 1.5M each (was 2.0M each, well under
PRD-014 SC-1 3M cap).
- [x] `npm run smoke` → `ALL IMAGES PASS`.
- [x] Manual scratch-dir probe: `init --image nightly` → `start` → `GET
/` returns HTTP 200; `GET /_app/immutable/entry/app.*.js` returns HTTP
200 with correct Content-Length.
- [ ] CI matrix (ubuntu/macos/windows) — pending.

Refs: PRD-031, EVID-036, PRD-014, PRD-030, #117.

Author: fedorovvvv

.forgeplan/evidence/EVID-036-drop-precompressed-assets-build-smoke-probe-measurements.md

@@ -0,0 +1,132 @@
+---
+depth: standard
+id: EVID-036
+kind: evidence
+last_modified_at: 2026-05-09T11:38:56.840360+00:00
+last_modified_by: claude-code/2.1.138
+links:
+- target: PRD-031
+  relation: informs
+status: active
+title: Drop precompressed assets — build/smoke/probe measurements
+---
+
+# EVID-036: Drop precompressed assets — build/smoke/probe measurements
+
+| Field | Value |
+|-------|-------|
+| Status | Draft |
+| Created | 2026-05-09 |
+| Valid Until | 2026-08-09 |
+| Target | PRD-031 |
+
+## Structured Fields
+
+evidence_type: measurement
+verdict: supports
+congruence_level: 3
+
+## Measurement
+
+Built `@forgeplan/web` after flipping `template/svelte.config.js` from
+`adapter()` to `adapter({ precompress: false })`. Measured against
+PRD-031 SC-1 / SC-2 / SC-3 / SC-4 / SC-5 directly on the post-build
+`dist/` and `dist-nightly/` directories and on a live server spawned
+from a scratch `init`. All commands executed in
+`/Users/nikitafedorov/Desktop/Projects/forgeplan-web` on
+`darwin 25.4.0` against the same `develop` branch the change was made on.
+
+Commands:
+
+```bash
+# SC-1: precompressed sibling count
+find dist dist-nightly -name '*.br' -o -name '*.gz' | wc -l
+
+# SC-2: per-image size
+du -sh dist dist-nightly
+
+# SC-3 + SC-4: live server probe
+SCRATCH=$(mktemp -d) && cd "$SCRATCH" && mkdir -p .forgeplan
+node /…/bin/forgeplan-web.mjs init -y --image nightly
+PORT=15998 node .forgeplan-web/index.js &
+curl -sI http://127.0.0.1:15998/_app/immutable/entry/app.DvKYkSiM.js
+curl -sI -H 'Accept-Encoding: br, gzip' \
+     http://127.0.0.1:15998/_app/immutable/entry/app.DvKYkSiM.js
+curl -s -o /dev/null -w "%{http_code}\n" http://127.0.0.1:15998/
+
+# SC-5: smoke
+npm run smoke
+```
+
+## Result
+
+| SC  | Metric | Before | After | Pass? |
+|-----|--------|--------|-------|-------|
+| SC-1 | precompressed siblings (`*.br` + `*.gz` under both `dist*/client/`) | 84 | **0** | ✅ |
+| SC-2 | `du -sh dist` | 2.0M | **1.5M** (-25%) | ✅ (< 3M cap) |
+| SC-2 | `du -sh dist-nightly` | 2.0M | **1.5M** (-25%) | ✅ (< 3M cap) |
+| SC-3 | `GET /` HTTP status (live server, scratch dir) | 200 | **200** | ✅ |
+| SC-4 | `/api/health` returns `ok` (smoke run) | ok | **ok** (project=shim) | ✅ |
+| SC-5 | `npm run smoke` exit code | 0 | **0** (`ALL IMAGES PASS`) | ✅ |
+
+`scripts/build.mjs` post-build report after the change:
+
+```
+[build] dist/ ready (1.40M, image=stable, features=0)
+[build] dist-nightly/ ready (1.40M, image=nightly, features=0)
+[build] done. images built: stable, nightly
+```
+
+Asset-compression probe (3047-byte `app.DvKYkSiM.js`):
+
+```
+--- without Accept-Encoding ---
+HTTP/1.1 200 OK
+Content-Length: 3047
+--- with Accept-Encoding: br, gzip ---
+HTTP/1.1 200 OK
+Content-Length: 3047
+```
+
+→ Sirv (adapter-node's static layer) serves uncompressed bytes when
+precompressed siblings are absent. There is **no** `Content-Encoding`
+header in either response. PRD-031 R-1 ("uncompressed first-load wire
+traffic") is therefore real, not just theoretical — but it lands on a
+loopback `127.0.0.1` connection in the canonical deployment, where the
+cost is negligible.
+
+## Interpretation
+
+Option A from issue #120 (disable `precompress` at adapter level) is
+implemented and verified end-to-end. All five PRD-031 success criteria
+pass on the first build. Tarball saving is ~520K per image (~1.04 MB
+total across both images) — exactly the figure forecast in the issue
+body. Live server still serves `GET /` HTTP 200 from a freshly
+initialised scratch dir; smoke matrix passes with `ALL IMAGES PASS`.
+
+The compression-probe result confirms the documented trade-off (R-1):
+sirv does NOT compress on-the-fly; clients receive uncompressed bytes.
+For local-first deployments (`127.0.0.1`, single user) this is
+negligible. If a future user reports slow first-load over a slow
+network, the mitigation path is documented in PRD-031 (flip
+`precompress` back to `true`, or wire an on-the-fly compression
+middleware in `template/src/hooks.server.ts`).
+
+This evidence supports activating PRD-031 with `R_eff > 0`.
+
+## Congruence Level Justification
+
+CL3 (same context, penalty 0.0): the measurement was executed against
+the exact build artifacts (`dist/`, `dist-nightly/`) and the exact bin
+script (`bin/forgeplan-web.mjs init … && node .forgeplan-web/index.js`)
+that PRD-031 governs, on the same `develop` branch the change was
+landed on. No proxy environment, no benchmark surrogate, no cached
+result — direct measurement on the surface the PRD describes.
+
+## Related Artifacts
+
+| Artifact | Relation |
+|----------|----------|
+| PRD-031 | informs |
+
+

.forgeplan/prds/PRD-014-reduce-dist-size-by-bundling-server-with-esbuild.md

@@ -228,3 +228,4 @@ Then используется именно /custom/path/forgeplan, а не де
 
 > **Next step**: создать RFC-013 с архитектурой бандла + эджкейсами.
 
+

.forgeplan/prds/PRD-030-feature-flag-and-image-system-promote-experimental-bundle-to-stable-default.md

@@ -300,3 +300,4 @@ And  neither directory contains a `node_modules/` subdirectory
 
 
 
+

.forgeplan/prds/PRD-031-drop-precompressed-client-assets-from-tarball.md

@@ -0,0 +1,210 @@
+---
+depth: standard
+id: PRD-031
+kind: prd
+last_modified_at: 2026-05-09T11:34:37.535120+00:00
+last_modified_by: claude-code/2.1.138
+links:
+- target: PRD-014
+  relation: based_on
+- target: PRD-030
+  relation: based_on
+status: active
+title: Drop precompressed client assets from tarball
+---
+
+---
+id: PRD-031
+title: "Drop precompressed client assets from tarball"
+status: Draft
+author: fedorovvvv
+created: 2026-05-09
+updated: 2026-05-09
+priority: P2
+depth: standard
+domain: general
+projectType: cli_tool
+stepsCompleted: []
+---
+
+# PRD-031: Drop precompressed client assets from tarball
+
+## Progress
+
+```
+Phase 0  ████████████████████████  1/1  (100%)
+─────────────────────────────────────────────────
+TOTAL                              1/1  (100%)
+```
+
+---
+
+## Executive Summary
+
+### Vision
+
+Ship a smaller `@forgeplan/web` tarball by removing 84 precompressed
+`*.br` / `*.gz` client siblings emitted by `@sveltejs/adapter-node` —
+saving ~520K per `dist*/` artifact (~29% of the experimental tarball)
+without measurably degrading the local-first user experience.
+
+### Problem
+
+`@sveltejs/adapter-node` defaults to `precompress: true`, so every
+`dist*/client/` carries 21 `.br` + 21 `.gz` files alongside the
+originals (~520K). The tarball ships **two** images (`dist/` for
+`stable`, `dist-nightly/` for `nightly`), so users pay the cost twice
+on every `npm install` / `npx` invocation.
+
+The runtime CPU savings precompressed assets buy (sirv serving the
+cached `.br`/`.gz` directly) are negligible for the canonical
+deployment shape: `npx @forgeplan/web start` on `127.0.0.1` for one
+local user, behind no CDN. Every user pays the install cost; almost
+nobody benefits from the runtime saving.
+
+**Impact**: ~520K × 2 dists = ~1.04 MB of dead weight on every
+install. After PR #117 minified the server bundle (1.4 MB → 683 KB),
+precompressed client assets are now the largest fixed cost in
+`dist-nightly/`.
+
+### Target Users
+
+| Persona | Описание | Ключевая боль |
+|---------|----------|---------------|
+| Local-first user | Runs `npx @forgeplan/web start` on `127.0.0.1` | Pays install cost (tarball download + extraction) for runtime savings they will never measure |
+| CI runner | Installs `@forgeplan/web` in throwaway environments | Slower cold installs because of inflated tarball |
+
+### Differentiators
+
+- Single-flag adapter change (Option A from issue #120) — minimal
+  blast radius, fully reversible by flipping `precompress` back to
+  `true`.
+- Aligned with `npx @forgeplan/web` philosophy: the package is
+  optimized for fast first-run on a developer machine, not for
+  high-throughput public-facing serving.
+
+---
+
+## Success Criteria
+
+| ID | Criterion | Metric | Current | Target | Timeframe | How to Measure |
+|----|-----------|--------|---------|--------|-----------|----------------|
+| SC-1 | Precompressed siblings absent from both shipped images | count of `*.br` + `*.gz` under `dist/client/` ∪ `dist-nightly/client/` after `npm run build` | 84 | 0 | This PR | `find dist dist-nightly -name '*.br' -o -name '*.gz' \| wc -l` |
+| SC-2 | Each tarball image stays under PRD-014 SC-1 cap | `du -sh dist*/` | 2.0M each | ≤ 1.5M each (target) / ≤ 3M (cap) | This PR | `du -sh dist dist-nightly` |
+| SC-3 | `start` still serves usable HTML | HTTP status of `GET /` after `npx @forgeplan/web start` | 200 | 200 | This PR | scratch-dir manual probe (see EVID) |
+| SC-4 | `/api/health` still returns `ok` | endpoint response | `{ ok: true }` | `{ ok: true }` | This PR | manual probe |
+| SC-5 | No regression in `npm run smoke` | smoke exit code | 0 | 0 | This PR | `npm run smoke` |
+
+---
+
+## Product Scope
+
+### MVP (In-Scope)
+
+- Pass `precompress: false` to `adapter()` in `template/svelte.config.js`.
+- Re-build both images (`dist/` + `dist-nightly/`) and verify the
+  precompressed siblings are gone.
+- Smoke test the bin script against a scratch dir end-to-end.
+- Document the trade-off (no precompressed cache; sirv now serves
+  uncompressed or compresses on-the-fly per `Accept-Encoding`).
+
+### Out of Scope
+
+- Lowering `DIST_*_MAX_BYTES` cap in `scripts/build.mjs` — separate
+  decision (PRD candidate).
+- RFC-013 graduation / legacy `dist/` removal — owned by PRD-014.
+- Adding a custom sirv compression middleware to recover the runtime
+  saving — only if a measured regression appears.
+
+### Growth Vision
+
+- If real users complain about uncompressed first-load wire size
+  over slow networks, revisit by either (a) flipping `precompress`
+  back on, or (b) wiring an on-the-fly compression middleware in
+  `template/src/hooks.server.ts`.
+
+---
+
+## User Journeys
+
+### Journey 1: Local-first user installs and starts
+
+**Цель пользователя**: Spin up the Forgeplan web viewer in a working repo.
+
+| Шаг | Действие пользователя | Ответ системы | Заметки |
+|-----|----------------------|---------------|---------|
+| 1 | `npx @forgeplan/web init -y` | Copies `dist/` (~1.5 MB target) into `.forgeplan-web/` | Smaller download than today |
+| 2 | `npx @forgeplan/web start` | Spawns node, listens on `127.0.0.1:5174` | No precompressed cache hit on assets |
+| 3 | Opens `http://127.0.0.1:5174` | Browser receives HTML; CSS/JS load | sirv serves assets uncompressed or via on-the-fly gzip per `Accept-Encoding` (TBD by smoke) |
+
+**Результат**: Same UX, smaller install footprint.
+
+### Journey 2: CI runner does throwaway install
+
+**Цель пользователя**: Validate Forgeplan integration in CI.
+
+| Шаг | Действие пользователя | Ответ системы | Заметки |
+|-----|----------------------|---------------|---------|
+| 1 | `npm i -g @forgeplan/web@latest` | Pulls smaller tarball, faster cold install | Saves ~1 MB per cold install across both images |
+| 2 | `forgeplan-web --version` | Prints version | Unaffected by this change |
+
+**Результат**: Faster CI cold start, no functional change.
+
+---
+
+## Functional Requirements
+
+| ID | Category | Priority | Requirement | Journey |
+|----|----------|----------|-------------|---------|
+| FR-001 | Build | Must | Build pipeline can emit `dist*/client/` directories with no `*.br` and no `*.gz` siblings | Journey 1, Journey 2 |
+| FR-002 | Runtime | Must | `start` command can serve an HTTP 200 on `GET /` and `/api/health` from a freshly initialised `.forgeplan-web/` that contains no precompressed siblings | Journey 1 |
+| FR-003 | Runtime | Should | Browser can render a usable page (CSS + JS load successfully) when the static-asset middleware has no precompressed siblings to fall back to | Journey 1 |
+
+---
+
+## Non-Functional Requirements
+
+| ID | Category | Requirement | Metric | Condition | Measurement |
+|----|----------|-------------|--------|-----------|-------------|
+| NFR-001 | Size | System shall emit each `dist*/` artifact below the existing cap | ≤ 3 MB | After `npm run build` for both `stable` and `nightly` images | `du -sh dist dist-nightly` + `scripts/build.mjs` cap assertion |
+| NFR-002 | Reversibility | System shall allow restoring precompressed assets by flipping a single boolean | 1 line in `template/svelte.config.js` | Forever | Diff inspection |
+| NFR-003 | Compatibility | System shall not regress `npm run smoke` exit code | exit code = 0 | Local + CI matrix (ubuntu/macos/windows) | `npm run smoke` |
+
+---
+
+## Risks & Mitigations
+
+| ID | Risk | Probability | Impact | Mitigation | Owner |
+|----|------|-------------|--------|------------|-------|
+| R-1 | Without precompressed siblings, sirv serves assets uncompressed → +700 KB first-load wire traffic for users on slow links | Medium | Low (local-first deployments dominate) | Document in README + this PRD; if real complaints surface, add an on-the-fly gzip middleware in `template/src/hooks.server.ts` or revert to `precompress: true` | fedorovvvv |
+| R-2 | A future custom hook in `template/` assumes `.br`/`.gz` siblings exist | Low | Medium | Grep template for `\.br` / `\.gz` references in this PR; none found at time of writing | fedorovvvv |
+| R-3 | Adapter-node behavior changes in a future major version, making `precompress: false` ineffective | Low | Low | Pinned via `template/package.json#dependencies`; bump-with-test is the existing process | fedorovvvv |
+
+---
+
+## Affected Files
+
+- `template/svelte.config.js` — flip `adapter()` → `adapter({ precompress: false })`.
+- `dist/`, `dist-nightly/` — regenerated by `npm run build` (no source-controlled change beyond the regenerated bundles).
+
+## Related Artifacts
+
+| Artifact | Relation | Status |
+|----------|----------|--------|
+| PRD-014 | parent (tarball-size cap SC-1) | active |
+| PRD-030 | parent (dual-image pipeline `stable` + `nightly`) | active |
+| PR #117 | predecessor (server bundle minify, 1.4 MB → 683 KB) | merged |
+| Issue #120 | source | open (this PRD closes it) |
+
+---
+
+> **Decision recorded (closes issue #120 acceptance criterion 1)**:
+> Option A — disable precompress at the adapter level. Rejected:
+> Option B (build.mjs strip — strictly more code for the same runtime
+> behaviour) and Option C (keep `.br` only — half the size win for
+> ~zero additional safety, since a sirv fallback regression would still
+> need investigation).
+
+
+
+

template/svelte.config.js

@@ -5,7 +5,7 @@ import { vitePreprocess } from '@sveltejs/vite-plugin-svelte';
 const config = {
   preprocess: vitePreprocess(),
   kit: {
-    adapter: adapter(),
+    adapter: adapter({ precompress: false }),
     files: {
       assets: 'static',
       lib: 'src/shared'