fix(nix): make the materialized .claude/ writable so the skills hook can write

[Soushi888]

Intent

Every nix develop entry and every CI job logs this, three times:

rsync: [Receiver] mkdir ".claude/skills/holochain" failed: Permission denied (13)
rsync error: error in file IO (code 11) at main.c(791)

It is non-fatal, but it is noise on every run and an active red herring when diagnosing real failures: in PR #130's failing e2e run the error appears immediately before launch-happ.mjs exited before ready, which invites the conclusion that they are related. They are not, and confirming that costs time each round.

Root cause

The shellHook materializes .claude/ out of the nix store:

rsync -a --delete ${./pai/claude}/ .claude/

rsync -a preserves permissions, and nix store paths are read-only, so .claude/ lands as mode 555. The agent-skills hook that runs a few lines later then cannot mkdir .claude/skills/holochain inside it.

.cursor/ is materialized the same way and already carries chmod -R u+w .cursor for exactly this reason. .claude/ was simply missing the equivalent line.

Change

One line, mirroring the .cursor/ treatment, placed after the rsync and before the skills hook, with a comment recording why it is load-bearing.

How to test

chmod -R u+w .claude && rm -rf .claude
nix develop --command true    # first run: materializes
nix develop --command true    # second run: this is the one that used to fail

Verified locally: .claude is now drwxr-xr-x, .claude/skills/holochain is created, and the second run produces no rsync output at all. Before the fix the second run reproduced the error every time.

Related

Clears one of the two prerequisites listed for reviewing #130 (the other being the 0.7 launcher timeout, which is genuinely unrelated and remains open).