Resolves F17 in docker-build-optimization-roadmap.
Root cause:
Gitea's app.ini ROOT_URL was http://localhost:3300/. Gitea bakes
ROOT_URL into the dist.tarball field of every published package's
metadata. Inside a Docker container, 'localhost' is the container
itself, not the host \u2014 so any 'pnpm install' that needed to fetch
a tarball would ECONNREFUSED, even though the registry metadata
itself was reachable via host.docker.internal.
Server-side fix (not in git, requires manual replication on each dev
machine; documented in roadmap \u00a73 A-pre-6):
- Edit /opt/homebrew/var/gitea/custom/conf/app.ini:
ROOT_URL = http://host.docker.internal:3300/
- brew services restart gitea
- sudo sh -c 'echo "127.0.0.1 host.docker.internal" >> /etc/hosts'
Repo-side fix (this commit):
- switch-network.sh: add host.docker.internal to NO_PROXY +
NPM_CONFIG_NOPROXY when NETWORK=corp. Required so host-side curl/
pnpm/npm bypass the corporate proxy (cso.proxy.att.com) when
resolving host.docker.internal. Without this, host installs fail
with the corp proxy's 'Unknown Host' 504 page.
Republished all 64 @bytelyst/* packages so tarball URLs reflect the
new ROOT_URL:
- .publish-manifest.json: 64 entries with new content hashes
- packages/*/package.json: 64 patch-version bumps
(auto-bumped by publish-outdated-packages.sh because previous
versions already existed in registry)
Verification:
curl http://localhost:3300/.../@bytelyst%2Ferrors | jq .dist.tarball
→ http://host.docker.internal:3300/.../errors-0.1.11.tgz (was localhost:3300)
workspace:* refs across all 64 packages: 0
Unblocks: A0-V on every pilot. Verified PASSING on learning_ai_clock:
backend cold build: 59.2 s
web cold build: 3:13 (193 s)
Both via Gitea registry, no docker-prep.sh tarballs needed.
Per user request: 'use the local Gitea and make sure all packages
in Gitea are up to date'.
Built all packages from learning_ai_common_plat/packages/* and ran
scripts/gitea/publish-outdated-packages.sh against the local Gitea
npm registry (http://localhost:3300/api/packages/bytelyst/npm/).
Manifest-based hash comparison flagged 4 packages whose built dist/
content has changed since the last published tarball:
@bytelyst/auth-ui 0.1.5 → 0.1.6
@bytelyst/config 0.1.7 → 0.1.8
@bytelyst/dashboard-shell 0.1.5 → 0.1.6
@bytelyst/mcp-client 0.1.0 → 0.1.1
All four bumped + published successfully. Remaining 60 packages
verified up-to-date. One package skipped by design:
@bytelyst/react-native-platform-sdk (RN — not in npm publish set)
Also incidentally fixed an mcp-client build break before this run:
stale dist/ + node_modules/.cache made tsc think MCPClient was
missing a 'log' property, even though the source had been correctly
refactored to use a private 'log: McpLogger' field. A clean
'rm -rf dist node_modules/.cache && pnpm build' resolved it; no
source changes needed.
Files updated:
- packages/auth-ui/package.json
- packages/config/package.json
- packages/dashboard-shell/package.json
- packages/mcp-client/package.json
- scripts/gitea/.publish-manifest.json (new content hashes)
After this commit, every published @bytelyst/* tarball in local
Gitea matches the source tree exactly.
Published 60 @bytelyst/* packages to local Gitea npm registry.
create-app skipped (private: true — internal scaffolding tool).
Token regenerated with full write:package scope to fix E401.
Root cause: tinypool worker teardown calls kill() which returns EPERM
in the act_runner host environment on Node.js v25.2.1. Tests pass but
the vitest process crashes during cleanup, causing CI failure.
Fix: --pool forks CLI flag on every package/service test script, plus
pool: 'forks' in all vitest.config.ts files. This uses child_process.fork()
worker management which handles termination cleanly.
60 package.json files updated, 10 vitest.config.ts files updated.
- Created .npmrc with @bytelyst scoped registry pointing to local Gitea
- Added publishConfig.registry to all 57 @bytelyst/* package.json files
- Created scripts/harden-publish-config.sh for future re-runs
- Prevents accidental publish to npmjs.org or corporate JFrog registry
