feat(tooling): shared-hooks dir so git worktree add installs node_modules on any host

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code

[patroza]

Aligns worktree setup across smart + the t3vm guest so a raw git worktree add installs node_modules on both — not only on the guest.

Why it differed

Husky sets core.hooksPath to a relative.husky/_, which doesn't exist in a freshly git worktree add-ed worktree, so git skips every hook there. On smart, node_modules only got installed through the T3 app / t3.jsonrunOnWorktreeCreate path; a raw CLI git worktree add came up empty. The guest worked only because t3-repos-setup installs a .git/hooks/post-checkout and clears the broken hooksPath.

Fix

install-git-hooks.mjs now points core.hooksPath at a shared absolute dir under the git common dir, which exists for every worktree (including brand-new ones):

• scripts/hooks/post-checkout (canonical worktree setup — pnpm install --frozen-lockfile --prefer-offline --hardlink, .worktreeinclude seeds) fires on git worktree add.
• Per-hook dispatchers delegate to the current worktree's checked-in .husky/<name>, so husky's lint-staged (pre-commit) + agent ship gate (pre-push) keep working.

Runs on every pnpm install, so both hosts converge with no host-specific installer. The canonical hook is ported from the guest's worktree-post-checkout.

Tests

• scripts/hooks/post-checkout.test.sh: no-op in the main worktree; fires on a fresh git worktree add; commit still runs husky pre-commit through the dispatcher.
• vp check clean; the git wiring was validated in an isolated throwaway repo (never mutating a live repo's hooksPath).

Note

Takes effect on the next pnpm install in any worktree (which repoints hooksPath). Ops follow-up: the guest t3-repos-setup post-checkout/hooksPath handling for t3code becomes redundant once this lands (it stays compatible — it only reinstalls an identical hook); I'll adjust guest.nix to defer to this to avoid drift.

🤖 Generated with Claude Code