pack-test

name	pack-test
description	Test the claude-second-brain npm package by packing it into a tarball and running it via npx. Exercises the full multi-brain scaffold flow end-to-end: scaffolds a brain, verifies config.toml, global skills, and local patches, then scaffolds a second brain to validate config upsert. Trigger phrases: /pack-test, test the package, test the CLI, test the tarball.
argument-hint	No arguments needed

Pack Test

Tests the npm package as it behaves when published. Uses npx to run from the local tarball — pnpm dlx doesn't support local tarball paths (it tries to resolve them as scoped package names). After publishing, pnpm dlx claude-second-brain from the registry works fine.

Multi-brain architecture: brains live at ~/.claude-second-brain/<name>/ and are registered in ~/.claude-second-brain/config.toml. Global skills (~/.claude/skills/brain-*) are plain SKILL.md files with a sibling .csb-version sidecar. They resolve paths at call time via the claude-second-brain path and claude-second-brain qmd subcommands — no paths are baked into the skills themselves.

Important: This test mutates ~/.claude-second-brain/ and ~/.claude/skills/brain-*. If the user has real brains registered, warn them before running — cleanup removes config.toml and the test brains, which does NOT delete their real brain folders but DOES remove their registry entries and overwrite their global skills.

Steps

Run each step in order. Stop and report if any step fails.

Step 1 — Pre-flight: back up existing state

# If the user has existing brains, these are affected:
ls -la ~/.claude-second-brain/ 2>/dev/null || echo "(no existing brains)"
cat ~/.claude-second-brain/config.toml 2>/dev/null || echo "(no existing config)"

If config.toml exists with real brains, back it up so the test cleanup doesn't lose them:

cp ~/.claude-second-brain/config.toml ~/.claude-second-brain/config.toml.bak 2>/dev/null || true

Step 2 — Pack

npm pack

Produces claude-second-brain-<version>.tgz in the repo root. Verify the tarball appears before continuing.

Step 3 — Scaffold first brain

npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz test-brain-a

Pass test-brain-a as the brain name argument to skip the interactive prompt. Non-TTY stdin skips the GitHub prompt. The brain is created at ~/.claude-second-brain/test-brain-a/.

Assertion: The CLI output should include cd ~/.claude-second-brain/test-brain-a in the Next steps section — not cd test-brain-a.

Step 4 — Verify single-brain scaffold

VAULT_A=~/.claude-second-brain/test-brain-a

# Brain is in the correct location (NOT in CWD)
[ -d "$VAULT_A" ] && echo "PASS: brain at $VAULT_A" || echo "FAIL: brain not at $VAULT_A"
[ ! -d "./test-brain-a" ] && echo "PASS: brain not in CWD" || echo "FAIL: brain leaked into CWD"

# .gitignore renamed from .gitignore.template
[ -f "$VAULT_A/.gitignore" ] && echo "PASS: .gitignore present" || echo "FAIL: .gitignore missing"

# qmd:setup actually created the sqlite db (not just a non-fatal warning)
[ -f "$VAULT_A/.qmd/index.sqlite" ] && echo "PASS: qmd:setup created index.sqlite" || echo "FAIL: qmd:setup did not create .qmd/index.sqlite"

# Wiki stubs, raw-sources dirs, local skills, scripts
ls $VAULT_A/wiki/ $VAULT_A/raw-sources/ $VAULT_A/.claude/skills/ $VAULT_A/scripts/qmd/

# Scripts use a relative .qmd path — no absolute path substitution
grep "\.qmd/index\.sqlite" $VAULT_A/scripts/qmd/setup.ts
grep "\.qmd/index\.sqlite" $VAULT_A/scripts/qmd/reindex.ts

# CLAUDE.md references the relative path too
grep "INDEX_PATH=\.qmd/index\.sqlite" $VAULT_A/CLAUDE.md | head -1

# No lingering tokens anywhere in the vault
! grep -r "__QMD_PATH__\|__CSB_CONFIG__\|__BRAIN_NAME__" $VAULT_A/ 2>/dev/null \
  && echo "PASS: no unpatched tokens" \
  || echo "FAIL: unpatched tokens found above"

Step 5 — Verify config.toml was created

CONFIG=~/.claude-second-brain/config.toml

[ -f "$CONFIG" ] && echo "PASS: config.toml created" || echo "FAIL: config.toml missing"
cat $CONFIG

# Assertions on content
grep -q 'default = "test-brain-a"' $CONFIG && echo "PASS: default brain set" || echo "FAIL: default not set"
grep -q 'name = "test-brain-a"' $CONFIG && echo "PASS: brain entry exists" || echo "FAIL: no brain entry"
grep -q "path = \"$HOME/.claude-second-brain/test-brain-a\"" $CONFIG && echo "PASS: path correct" || echo "FAIL: path wrong"
grep -q "qmd_index = \"$HOME/.claude-second-brain/test-brain-a/.qmd/index.sqlite\"" $CONFIG && echo "PASS: qmd_index correct" || echo "FAIL: qmd_index wrong"

Step 6 — Verify global skills are CLI-backed

for skill in brain-ingest brain-search brain-refresh; do
  FILE=~/.claude/skills/$skill/SKILL.md
  SIDECAR=~/.claude/skills/$skill/.csb-version
  [ -f "$FILE" ] || { echo "FAIL: $FILE missing"; continue; }

  # Must reference the `claude-second-brain` CLI for path / qmd resolution
  grep -q "claude-second-brain path" $FILE && echo "PASS: $skill uses \`claude-second-brain path\`" || echo "FAIL: $skill missing CLI path resolution"
  grep -q "claude-second-brain qmd" $FILE && echo "PASS: $skill uses \`claude-second-brain qmd\`" || echo "FAIL: $skill missing CLI qmd proxy"

  # No lingering tokens, no baked absolute paths
  ! grep -q "__QMD_PATH__\|__CSB_CONFIG__\|INDEX_PATH=\$QMD_INDEX\|INDEX_PATH=/" $FILE && echo "PASS: $skill clean" || echo "FAIL: $skill still has legacy substitutions"

  # Version sidecar exists and matches package version
  [ -f "$SIDECAR" ] && echo "PASS: $skill .csb-version present" || echo "FAIL: $skill missing .csb-version"
  PKG_VER=$(node -p "require('./package.json').version")
  grep -q "\"version\": \"$PKG_VER\"" $SIDECAR && echo "PASS: $skill version matches package" || echo "FAIL: $skill version mismatch"
done

Step 7 — Scaffold second brain (multi-brain upsert test)

npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz test-brain-b

Step 8 — Verify config.toml upsert

CONFIG=~/.claude-second-brain/config.toml
cat $CONFIG

# default should STILL be test-brain-a (first brain stays default)
grep -q 'default = "test-brain-a"' $CONFIG && echo "PASS: default unchanged" || echo "FAIL: default was overwritten"

# Both brains should be registered
grep -c '^\[\[brains\]\]' $CONFIG
[ "$(grep -c '^\[\[brains\]\]' $CONFIG)" = "2" ] && echo "PASS: 2 brains registered" || echo "FAIL: expected 2 brain entries"

grep -q 'name = "test-brain-a"' $CONFIG && echo "PASS: brain-a entry exists" || echo "FAIL: brain-a missing"
grep -q 'name = "test-brain-b"' $CONFIG && echo "PASS: brain-b entry exists" || echo "FAIL: brain-b missing"

Step 9 — Verify CLI subcommands resolve the default brain

RESOLVED_ROOT=$(npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz path)
[ "$RESOLVED_ROOT" = "$HOME/.claude-second-brain/test-brain-a" ] && echo "PASS: path resolves default brain" || echo "FAIL: path returned '$RESOLVED_ROOT'"

RESOLVED_QMD=$(npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz path --qmd)
[ "$RESOLVED_QMD" = "$HOME/.claude-second-brain/test-brain-a/.qmd/index.sqlite" ] && echo "PASS: path --qmd resolves default brain" || echo "FAIL: path --qmd returned '$RESOLVED_QMD'"

RESOLVED_B=$(npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz path --brain test-brain-b)
[ "$RESOLVED_B" = "$HOME/.claude-second-brain/test-brain-b" ] && echo "PASS: --brain overrides default" || echo "FAIL: --brain returned '$RESOLVED_B'"

Step 9b — Verify new CLI surface (v1.1+)

CSB="npx -y claude-second-brain-$(node -p "require('./package.json').version").tgz"

# `csb use` — switch default to brain-b, then back
$CSB use test-brain-b > /dev/null && echo "PASS: csb use test-brain-b" || echo "FAIL: csb use test-brain-b"
grep -q 'default = "test-brain-b"' ~/.claude-second-brain/config.toml && echo "PASS: default switched to brain-b" || echo "FAIL: default did not switch"
$CSB use test-brain-a > /dev/null && echo "PASS: csb use test-brain-a (switch back)" || echo "FAIL: csb use test-brain-a"
grep -q 'default = "test-brain-a"' ~/.claude-second-brain/config.toml && echo "PASS: default back to brain-a" || echo "FAIL: default did not revert"

# `csb path qmd` (positional) should match legacy `csb path --qmd`
POS_QMD=$($CSB path qmd)
LEGACY_QMD=$($CSB path --qmd 2>/dev/null)
[ "$POS_QMD" = "$LEGACY_QMD" ] && [ "$POS_QMD" = "$HOME/.claude-second-brain/test-brain-a/.qmd/index.sqlite" ] \
  && echo "PASS: csb path qmd (positional = legacy flag)" \
  || echo "FAIL: positional='$POS_QMD' legacy='$LEGACY_QMD'"

# `csb path config` (positional) prints central config path
[ "$($CSB path config)" = "$HOME/.claude-second-brain/config.toml" ] && echo "PASS: csb path config" || echo "FAIL: csb path config"

# `csb exec -- pwd` runs inside the resolved brain's root
PWD_OUT=$($CSB exec -- pwd)
[ "$PWD_OUT" = "$HOME/.claude-second-brain/test-brain-a" ] && echo "PASS: csb exec -- pwd" || echo "FAIL: csb exec -- pwd returned '$PWD_OUT'"

# `csb doctor` exits 0 on a fresh scaffold
$CSB doctor > /dev/null 2>&1 && echo "PASS: csb doctor exits 0" || echo "FAIL: csb doctor exited non-zero"

# Per-subcommand --help routes to topic help instead of falling through to createBrain
$CSB rm --help 2>&1 | grep -q "csb rm —" && echo "PASS: csb rm --help" || echo "FAIL: csb rm --help fell through"
$CSB path --help 2>&1 | grep -q "csb path —" && echo "PASS: csb path --help" || echo "FAIL: csb path --help fell through"

Step 10 — Final checklist

Item	Expected	Pass?
~/.claude-second-brain/test-brain-a/	Present
~/.claude-second-brain/test-brain-b/	Present
Brain NOT leaked into CWD	./test-brain-a does not exist
.gitignore (not .gitignore.template)	Present in both brains
wiki/{index,log,overview}.md, sources/, qa/	Present in both brains
raw-sources/{articles,pdfs,personal}/	Present in both brains
.claude/skills/ — 5 subdirs	brain-ingest, brain-search, brain-refresh, brain-rebuild, lint
scripts/qmd/{setup,reindex}.ts	Present in both brains, DB hardcoded to .qmd/index.sqlite relative
CLAUDE.md, mise.toml, package.json	Present in both brains
CLAUDE.md INDEX_PATH	Relative INDEX_PATH=.qmd/index.sqlite (no absolute substitution)
No unpatched tokens	No __QMD_PATH__, __CSB_CONFIG__, __BRAIN_NAME__ anywhere in vault
~/.claude-second-brain/config.toml	Present with default = "test-brain-a" and 2 [[brains]] entries
Global skills (~/.claude/skills/brain-{ingest,search,refresh}/SKILL.md)	Present; reference claude-second-brain path + claude-second-brain qmd; .csb-version sidecar matches package version
csb use <name>	Rewrites default = "…" in config.toml
csb path qmd (positional)	Equal to csb path --qmd
csb path config (positional)	Prints ~/.claude-second-brain/config.toml
csb exec -- pwd	Prints the resolved brain's directory
csb doctor	Exits 0 on a fresh scaffold
csb <subcommand> --help	Each subcommand routes to its own topic help (no fall-through to createBrain)

Step 11 — Clean up

rm -rf ~/.claude-second-brain/test-brain-a
rm -rf ~/.claude-second-brain/test-brain-b
rm -f ~/.claude-second-brain/config.toml
rm -rf ~/.claude/skills/brain-ingest ~/.claude/skills/brain-search ~/.claude/skills/brain-refresh
rm claude-second-brain-$(node -p "require('./package.json').version").tgz

# Restore backed-up config if it existed
[ -f ~/.claude-second-brain/config.toml.bak ] && mv ~/.claude-second-brain/config.toml.bak ~/.claude-second-brain/config.toml

# Remove the parent dir only if empty
rmdir ~/.claude-second-brain/ 2>/dev/null || true

# Clean up GitHub test repos (if any were created — normally no, since non-TTY skips the prompt)
gh repo delete test-brain-a --yes 2>/dev/null || true
gh repo delete test-brain-b --yes 2>/dev/null || true

Note: If the user had pre-existing brain-ingest / brain-search / brain-refresh global skills from a real vault, they were overwritten by this test. They need to re-run npx claude-second-brain or manually re-scaffold a brain to restore them. The config backup in Step 1 protects their brain registry but not their global skills.

What failure means

• .gitignore.template present instead of .gitignore — npm stripped .gitignore from the tarball but the rename in bin/create.js didn't fire; check the try/catch rename block
• Brain leaked into CWD — targetDir in main() still uses process.cwd(); check bin/create.js uses join(CSB_ROOT, targetName)
• Missing skill dirs — template copy failed or .claude/ hidden dir wasn't included; check npm pack --dry-run output
• npx fails to run the bin — shebang or bin field in package.json is wrong; check #!/usr/bin/env node on line 1 of bin/create.js
• __QMD_PATH__ or __CSB_CONFIG__ still in any file — those substitutions were removed; any hit means a template file still has a legacy placeholder that should be replaced with a relative path or CLI invocation
• Global skills still reference $QMD_INDEX — the old Brain Discovery preamble wasn't dropped; regenerate from the new template/.claude/skills/brain-*/SKILL.md sources
• .csb-version missing — installGlobalSkills() didn't write the sidecar; check the sidecar write in bin/create.js
• config.toml missing — writeConfig() not called or failed; check the call at the end of main()
• default changed after second brain was created — writeConfig() upsert logic is overwriting default; it should only set default when the config file doesn't exist yet
• Only one [[brains]] entry after second scaffold — writeConfig() upsert is replacing instead of appending; check the split/filter/join logic
• claude-second-brain path returns wrong brain — parseConfig regex failing to pick up default = … (or legacy active = …); inspect the regex
• Note: pnpm dlx cannot run local tarballs — use npx for this test. After publishing, pnpm dlx claude-second-brain from the registry works fine.