post-mortem: mac-mini onboarding — 12 issues found

investigations/2026-06-01-mac-mini-onboarding.md

@@ -0,0 +1,185 @@
+# mac-mini onboarding — post-mortem & lessons learned
+
+Onboarded the `mac-mini` workstation (macOS Sequoia, arm64) into the hubris
+homelab context system with the `--with-hermes` profile. What follows is a
+chronological recap of every hitch, the fix, and the systemic improvements
+needed so the next workstation takes 5 min instead of an hour.
+
+## Session log
+
+### Step 1 — clone + symlink
+- Manually cloned `git.hubris.network/dtoro/Homelab-Docs` to `/Users/dtoro/Homelab-Docs`.
+- Created `/opt/homelab-context` → `/Users/dtoro/Homelab-Docs` symlink.
+- **Lesson:** bootstrap.sh was designed to do this from scratch, but we'd
+  already cloned by hand. The bootstrap's `clone exists; pulling` path handled
+  it gracefully.
+
+### Step 2 — hostname mismatch
+- `scutil --get LocalHostName` → `Davids-Mac-mini`
+- `hostname -s` → `Mac`
+- Inventory file: `hosts/mac-mini.yaml`
+- **Fix:** `sudo scutil --set LocalHostName mac-mini && sudo scutil --set HostName mac-mini`
+- **Lesson:** The bootstrap and `homelab whoami` use different hostname
+  resolution. Bootstrap uses `scutil --get LocalHostName` (correct on macOS),
+  but the `homelab` CLI binary uses `hostname -s`. Both need to match the
+  inventory key. On a fresh macOS machine, neither does.
+
+### Step 3 — bootstrap dependencies
+- pyyaml was missing → `pip install pyyaml`
+- age and sops were missing → `brew install age sops`
+- Netbird was already installed and connected ✓
+- **Lesson:** The bootstrap preflight handles these, but only if running
+  `bootstrap.sh` from the start. Since we ran it after manual setup, some
+  steps (netbird install) were correctly skipped as already-present.
+
+### Step 4 — full bootstrap with `--with-mcp --with-hermes`
+- Ran `sudo HOMELAB_GITEA_TOKEN=... bash bootstrap.sh --with-mcp --with-hermes`
+- Age key issued ✓
+- Launchd sync timer installed ✓
+- Goose binary installed ✓
+- Hermes CLI linked ✓
+- MCP config merged ✓
+- `refresh-creds` skipped (not yet a recipient) ⚠️
+- Cosmetics: `chown: dtoro: illegal group name` at the end (benign, macOS
+  group-naming quirk)
+
+### Step 5 — finalize from hubris
+- Ran `homelab client add mac-mini --finalize-pubkey <age...> --with-hermes` on
+  hubris
+- Push failed: `[rejected] main -> main (fetch first)` — hubris clone was
+  stale, bootstrap had already pushed from mac-mini
+- **Fix:** `git pull --rebase && git push` on hubris
+- **Lesson:** bootstrap pushes remote changes before hubris can finalize,
+  creating a race. The `homelab client add --finalize-pubkey` command should
+  pull before committing/pushing.
+
+### Step 6 — sops couldn't find the age key
+- `homelab secret hello` failed because sops looks in
+  `/Users/dtoro/.ssh/id_rsa` etc. by default, not `/etc/age/key.txt`
+- The `homelab` CLI re-execs via `sudo -E env SOPS_AGE_KEY_FILE=... sops ...`,
+  but this requires passwordless sudo and the correct env var passthrough
+- **Fix:**
+  1. Added NOPASSWD sudo rules
+  2. Eventually `SOPS_AGE_KEY` env with the raw key content worked directly
+- **Lesson:** Document the explicit `SOPS_AGE_KEY_FILE` incantation in
+  agent-enrollment troubleshooting. New clients can't assume `homelab secret`
+  works out of the gate — the sudo re-exec chain has permission pitfall.
+
+### Step 7 — OpenRouter key was a placeholder
+- `secrets/openrouter-api-key.yaml` contained
+  `api_key: PLACEHOLDER_REPLACE_WITH_REAL_OPENROUTER_KEY`
+- User ran `sops` on hubris, but got the same error (age key not found on
+  hubris either — `/root/.config/sops/age/keys.txt` didn't exist)
+- **Fix:** `SOPS_AGE_KEY_FILE=/etc/age/key.txt sops ...` on hubris.
+  Later: the user pasted the real key, but the sops file showed
+  `sk-or-...5c55` — the literal content was truncated with ellipsis.
+
+### Step 8 — editor loaded the wrong data
+- Neovim on the system is configured with `clipboard+=unnamedplus`, which
+  points `*` and `+` registers to the macOS clipboard manager rather than
+  X11. When editing SOPS files, this caused the **system clipboard** to be
+  pasted instead of the actual ciphertext.
+- This wasn't diagnosed during the session — the sops file would load empty
+  or show the wrong content because the editor's idea of "paste" was
+  disconnected from what sops expected.
+- **Fix:** Run `sops` with `EDITOR=nano` or another editor that doesn't
+  hijack OS clipboards:
+  ```bash
+  EDITOR=nano SOPS_AGE_KEY_FILE=/etc/age/key.txt sops secrets/openrouter-api-key.yaml
+  ```
+- **Lesson:** Add a strong warning to `hermes-agent.md` / `agent-enrollment.md`:
+  macOS neovim with `clipboard+=unnamedplus` silently breaks sops editing
+  because the paste register reads from the system clipboard instead of the
+  sops-managed buffer. Use `EDITOR=nano` or `EDITOR=vim` when running sops
+  interactively. Alternatively, override the clipboard option with
+  `EDITOR='nvim -c "set clipboard="'`.
+- Also useful for the troubleshooting table in `agent-enrollment.md` under a
+  new row: "sops file loads empty / wrong content on macOS"
+
+### Step 9 — model doesn't support tool use
+- Goose config pinned `nousresearch/hermes-4-405b` via OpenRouter
+- Error: `No endpoints found that support tool use`
+- **Fix:** Switched to `deepseek/deepseek-v4-flash` in
+  `~/.config/goose/config.yaml`
+- Also updated `operations/hermes-agent.md` with the correct model
+- **Lesson:** The default model in `bootstrap.sh` and `hermes-agent.md` was
+  never validated on OpenRouter for tool-use capability. Need to either:
+  (a) Pin a model known to work (`deepseek/deepseek-v4-flash`)
+  (b) Or have the bootstrap probe OpenRouter at install time
+
+### Step 10 — MCP homelab extension returns 404
+- Goose's homelab extension configured as `streamable_http` at
+  `https://mcp.hubris.network/mcp` returns HTTP 404
+- The actual MCP server runs on SSE (`/sse`), not streamable_http
+- **Known follow-up** from `hermes-agent.md` — server needs migration
+- **Workaround:** The agent works fine via the built-in `developer` extension
+  (shell + file) and the `homelab` CLI
+
+### Step 11 — git credential dance
+Several layers of issues with git authentication:
+
+1. **Directory permissions:** `/etc/homelab-context` was `0700 root` — dtoro
+   couldn't access the credential file, and git's `store` helper couldn't
+   create its `.lock` file.
+   - **Fix 1a:** `sudo chmod 755 /etc/homelab-context`
+   - **Fix 1b:** `sudo chown -R dtoro:staff /etc/homelab-context`
+
+2. **macOS built-in osxkeychain:** Apple's git has `osxkeychain` compiled in
+   as a default credential helper. Even after configuring the `store` helper,
+   git calls osxkeychain after every successful `get`, which prompts for
+   macOS keychain access.
+   - `credential.helper = ""` does NOT disable the compiled-in default
+   - The `store` helper's `.lock` file in `/etc/homelab-context/` also failed
+     silently, causing fallback behaviour
+   - **Fix:** Embed the credential in the remote URL directly:
+     `https://dtoro:TOKEN@git.hubris.network/dtoro/Homelab-Docs.git`
+   - This bypasses all credential helpers entirely
+
+3. **refresh-creds:** The credential is now managed by the remote URL.
+   If the PAT is rotated, `homelab refresh-creds` won't update the URL.
+   - **Workaround:** Run `git remote set-url origin` after `refresh-creds`
+
+### Step 12 — `homelab` CLI hostname detection
+- `homelab whoami` failed with `no hosts/Mac.yaml` because the CLI uses
+  `hostname -s` (returning `Mac`) while the bootstrap uses
+  `scutil --get LocalHostName` (which we fixed to `mac-mini`)
+- **Fix:** Set both `LocalHostName` and `HostName` via `scutil`
+- **Lesson:** The homelab CLI on macOS should prefer `scutil --get
+  LocalHostName` like the bootstrap does, or at least try both and pick the
+  one that matches an inventory key. This is a CLI bug.
+
+## Improvement backlog
+
+### High priority
+
+| # | Issue | Fix |
+|---|-------|-----|
+| 1 | Homelab CLI uses `hostname -s` on macOS; should use `scutil --get LocalHostName` to match bootstrap | Patch `bin/homelab` to try `scutil` first, fall back to `hostname -s` |
+| 2 | `homelab client add --finalize-pubkey` on hubris fails to push when remote is ahead (bootstrap pushes first) | `pull --rebase` before committing in `cmd_client_add` |
+| 3 | Default Goose model (`nousresearch/hermes-4-405b`) doesn't support tool use on OpenRouter | Update `bootstrap.sh` default and `hermes-agent.md` to `deepseek/deepseek-v4-flash` |
+| 4 | MCP server needs streamable_http migration | Follow-up #1 from `hermes-agent.md` — FastMCP `transport="sse"` → `transport="streamable_http"` |
+| 5 | `homelab refresh-creds` should also update embedded remote URLs | In `cmd_refresh_creds`, after writing the credential file, update any remote URL that has an embedded credential |
+
+### Medium priority
+
+| # | Issue | Fix |
+|---|-------|-----|
+| 6 | macOS neovim with `clipboard+=unnamedplus` breaks sops editing | Add troubleshooting row to `agent-enrollment.md`; recommend `EDITOR=nano` |
+| 7 | `/etc/homelab-context` directory with 0700 root permissions blocks non-root git | Change to 755 in `bootstrap.sh` on all OSes (or at least macOS) |
+| 8 | Bootstrap `chown` at end gives `illegal group name` on macOS | Fix group detection in bootstrap.sh for macOS |
+| 9 | `homelab secret` re-exec via sudo chain is fragile — needs passwordless sudo or TTY | Consider `SOPS_AGE_KEY` env-var fallback path in the CLI itself |
+
+### Low priority
+
+| # | Issue | Fix |
+|---|-------|-----|
+| 10 | Onboarding this machine revealed no `HERMES.md` file in the repo | Create the persona file for the Goose hints symlink |
+| 11 | Git credential `store` helper lock mechanism is OS-sensitive and fragile | Document embedded-URL pattern as the preferred approach for workstations |
+| 12 | AGENTS.md references `/opt/homelab-context/hosts/<hostname>.yaml` but the canonical path is now a symlink | Verify the path resolution edge cases |
+
+## Changelog
+
+### 2026-06-01 — initial post-mortem
+Captured the full onboarding of mac-mini into the hubris homelab context
+system, including the hostname mismatch, sudo/sops credential chain, model
+selection, git credential gymnastics, and editor clipboard interference.