Rewrite PIA VPN as multi-container bridge architecture
All checks were successful
Check Flake / check-flake (push) Successful in 4m44s

Replace the single VPN container (veth pair, host-side auth scripts) with a
multi-container setup on a shared bridge network:

- Dedicated VPN container handles all PIA auth, WireGuard config, NAT, and
  optional port forwarding DNAT
- Service containers default-route through VPN container (leak-proof by topology)
- Host runs tinyproxy on bridge for PIA API bootstrap before WG is up
- WG interface is still created in host netns and moved into VPN container
  namespace
- Monthly renewal to ensure that connection stays up (PIA allows connections to
  last up to 2 months)
- Drop OpenVPN support entirely
This commit is contained in:
2026-02-22 23:10:25 -08:00
committed by Zuckerberg
parent 684851d641
commit a0fcacdcf9
13 changed files with 979 additions and 885 deletions

View File

@@ -86,8 +86,16 @@ When adding or removing a web-facing service, update both:
- Don't use `nix build --dry-run` unless you only need evaluation — it skips the actual build
- Avoid `2>&1` on nix commands — it can cause error output to be missed
## Git Worktree Requirement
## Git Worktrees
When instructed to work in a git worktree (e.g., via `isolation: "worktree"` or told to use a worktree), you **MUST** do so. If you are unable to create or use a git worktree, you **MUST** stop work immediately and report the failure to the user. Do not fall back to working in the main working tree.
When the user asks you to "start a worktree" or work in a worktree, **do not create one manually** with `git worktree add`. Instead, tell the user to start a new session with:
```bash
claude --worktree <name>
```
This is the built-in Claude Code worktree workflow. It creates the worktree at `.claude/worktrees/<name>/` with a branch `worktree-<name>` and starts a new Claude session inside it. Cleanup is handled automatically on exit.
When instructed to work in a git worktree (e.g., via `isolation: "worktree"` on a subagent), you **MUST** do so. If you are unable to create or use a git worktree, you **MUST** stop work immediately and report the failure to the user. Do not fall back to working in the main working tree.
When applying work from a git worktree back to the main branch, commit in the worktree first, then use `git cherry-pick` from the main working tree to bring the commit over. Do not use `git checkout` or `git apply` to copy files directly. Do **not** automatically apply worktree work to the main branch — always ask the user for approval first.