[BUG] supercode init exits silently during 'loading conversation' hook — prompt never opens (v0.1.90, paid tier)

Describe the bug

supercode init (CLI v0.1.90) hangs in "loading conversation" during startup and silently exits after ~6 ticks (~5s) with the message thanks for being here · see you next time. The interactive prompt never opens — stdin is never read. No error is emitted to stdout, stderr, or to any log file. There is no --verbose / --debug flag to inspect what is hanging.

I reproduced this 3 times with 3 different configurations and the behavior is identical every time. Same final 5 lines of output regardless of input.

To Reproduce

Steps:

1. Authenticate once: supercode login (successful — Better-Auth device flow, paid tier, account Jose Isaias)
2. Run supercode init from any directory
3. Observe splash → auth "Welcome, Jose Isaias" → "paid tier" → SUPERCODE ai chat · <model> blocks → 6 ticks of loading conversation → process exits with thanks for being here · see you next time
4. Prompt [chat] > never opens, no input is accepted

The same outcome happens with:

• supercode init (default config exists at ~/.config/supercode/cli-config.json)
• supercode init (after rm ~/.config/supercode/cli-config.json)
• supercode init --resume="" (forced new conversation)
• bash -c 'supercode init <<< "test"' (stdin pre-populated)

Expected behavior

The CLI should open the [chat] > prompt and wait for real user input. If the conversation-load hook times out, it should fall back to a fresh conversation and show a visible error (e.g. failed to resume conversation 'abc123', starting fresh) rather than silently terminate with a "thanks for being here" farewell.

Screenshots / Logs

The 3 attempts produce byte-identical tails. Final lines of every run:

┃ supercode · chat · deepseek-v4-flash ───────────────────── ready · type to chat

┌─ directory ──────────────────────────────────────────────────────────────┐
│   workspace · supercli-inspect                                         │
│   path     · /tmp/supercli-inspect                                     │
│   files    · 2                                                         │
└──────────────────────────────────────────────────────────────────────────┘

  hint · /model to switch  · Ctrl+V voice  · /help for commands  · Tab to cycle mode

▏ authenticating ▎ authenticating ▍ authenticating ▌ authenticating ▋ authenticating ◆ Welcome, Jose Isaias
▏ loading conversation ▎ loading conversation ▍ loading conversation ▌ loading conversation ▋ loading conversation▌ [chat] > Ask anything... "Fix broken tests"
  ◇  thanks for being here  ◇
     see you next time · supercode ◆

Note the prompt placeholder [chat] > Ask anything... "Fix broken tests" is rendered right next to the last loading conversation tick — the prompt is shown but the process exits immediately without ever reading stdin.

Mid-session state on each run:

• OK: Welcome, Jose Isaias + paid tier
• OK: server assigns model (saw kimi-k2-6 and deepseek-v4-flash across runs)
• FAIL: loading conversation × 6 ticks → process dies
• Note: ~/.config/supercode/cli-config.json shows "model": "deepseek-v4-flash" after exit, but the model is server-assigned, not local.

Environment

• OS: Linux 6.8.0-134-generic (Ubuntu 22.04 container)
• Supercode version: 0.1.90 (installed via npm install -g supercode-cli, today, 2026-07-25)
• Shell: bash
• Node: v22.22.3 (/root/.hermes/node/bin/node)
• Bun: not installed (used npm — same package)
• Auth: Better-Auth device flow, paid tier, account Jose Isaias
• Server: default https://supercode-8w7e.onrender.com (not overridden)

Additional context

• No --verbose, --debug, -v, or --log-level flag exists on supercode, supercode init, supercode login, or supercode skill. Verified each via --help.
• No log file produced anywhere under ~/.supercode/, ~/.better-auth/, ~/.cache/, ~/.local/share/, /tmp, or /var/log. Only files left behind are:
  • ~/.better-auth/token.json (Bearer token, expires 2026-08-01)
  • ~/.config/supercode/cli-config.json (recreated on each run with server-assigned model)
• The "thanks for being here" message is a deliberate farewell — the CLI is choosing to exit, not crashing. Suspect the conversation-load hook either:
  • hits a silent timeout against the Render free tier (supercode-8w7e.onrender.com), or
  • is a bug in v0.1.90 where the load hook never resolves when no --resume is supplied and the server has no prior conversation
• Likely related: in my previous session I switched to glm-5.1 via the in-CLI /model command. The CLI may be trying to resume that conversation and failing.

Suggested fix (similar to issue #205):

• Add a --verbose or --debug flag to surface the actual error from the conversation-load hook
• When the hook fails/times out, fall back to a fresh conversation and log a visible warning
• Don't exit with a "thanks for being here" farewell when the user has never interacted yet — that's misleading

Start at the supercode init entry point and trace the loading conversation hook, including the --resume path and configuration at ~/.config/supercode/cli-config.json. Reproduce with the listed supercode init variants and inspect how timeout or load failure is handled. Done means the prompt remains interactive, or a visible fallback error is shown instead of an immediate farewell.