70 lines
5.6 KiB
Markdown
70 lines
5.6 KiB
Markdown
# Fix: Kind 30078 `d=agent_config` Publish Gaps
|
|
|
|
## Problem
|
|
|
|
The agent stores its admin pubkey and DM protocol in a NIP-44 self-encrypted kind 30078 replaceable event (`d=agent_config`). This event is critical for the `--nsec`-only startup path (used by systemd services) to recover the admin identity.
|
|
|
|
**Multiple startup paths fail to publish this event**, causing:
|
|
- Agents installed via wizard "Install Service" to lose their admin pubkey on restart
|
|
- Agents recovering on new servers to use stale/old admin pubkeys from relays
|
|
- Agents with genesis files to silently ignore newer kind 30078 values
|
|
|
|
## Affected Files
|
|
|
|
- `src/setup_wizard.c` — wizard flows
|
|
- `src/main.c` — main startup logic
|
|
|
|
## Fixes
|
|
|
|
### Fix A: `new_agent_flow` — publish before service install
|
|
**File:** `src/setup_wizard.c` ~line 2184
|
|
**Problem:** When user chooses "Install systemd service" in new agent flow, `persist_runtime_config_to_nostr_wizard_online` is never called. The service starts with `--nsec` only and has no kind 30078 to recover from.
|
|
**Fix:** Call `persist_runtime_config_to_nostr_wizard_online(cfg)` before `install_system_service_with_dedicated_user()`. Fail the install if publish fails (same pattern as existing_agent_flow).
|
|
|
|
### Fix B: `new_agent_flow` — publish before "boot now" return
|
|
**File:** `src/setup_wizard.c` ~line 2177
|
|
**Problem:** When user chooses "Boot now", the wizard returns `SETUP_WIZARD_RC_BOOTSTRAP` and relies on `main()` to publish later. This works but is fragile — if the bootstrap publish in `main()` fails, the kind 30078 is never created.
|
|
**Fix:** Call `persist_runtime_config_to_nostr_wizard_online(cfg)` before returning 0. This is belt-and-suspenders — `main()` will also publish, but the wizard ensures it happens at least once. If the wizard publish fails, log a warning but continue (non-fatal since main will retry).
|
|
|
|
### Fix C: `existing_agent_flow` — always publish before service install
|
|
**File:** `src/setup_wizard.c` ~line 2368
|
|
**Problem:** `persist_runtime_config_to_nostr_wizard_online` is only called `if (config_changed)`. If the user accepts the recovered config as-is, the kind 30078 is not re-published. The event may have been lost from relays.
|
|
**Fix:** Remove the `config_changed` gate. Always call `persist_runtime_config_to_nostr_wizard_online(cfg)` before installing the service. Kind 30078 is a replaceable event, so re-publishing is safe and idempotent.
|
|
|
|
### Fix D: `existing_agent_flow` — always return BOOTSTRAP
|
|
**File:** `src/setup_wizard.c` ~line 2364
|
|
**Problem:** When user chooses "Boot now" and `config_changed == 0`, the flow returns `SETUP_WIZARD_RC_EXISTING` which means `bootstrap_mode = 0` in `main()`. If `first_run = 0` (kind 10002 exists), `persist_runtime_config_to_nostr` is skipped.
|
|
**Fix:** Always return `SETUP_WIZARD_RC_BOOTSTRAP` from the "boot now" path, regardless of `config_changed`. This ensures `main()` always re-publishes the kind 30078. The wizard has all the config in memory — it should always be treated as authoritative.
|
|
|
|
### Fix E: `persist_runtime_config_to_nostr_wizard_online` — ensure relay delivery
|
|
**File:** `src/setup_wizard.c` ~line 1159
|
|
**Problem:** The function publishes the event and immediately calls `nostr_handler_cleanup()`. The event may not have been delivered to relays yet (fire-and-forget).
|
|
**Fix:** Add a brief poll loop (e.g., 2-3 seconds of `nostr_handler_poll()`) after the publish call and before `nostr_handler_cleanup()` to give the event time to propagate. This matches the pattern used elsewhere for relay operations.
|
|
|
|
### Fix F: `main()` — always fetch kind 30078 and compare
|
|
**File:** `src/main.c` ~line 847 (`recover_missing_runtime_config_from_nostr`)
|
|
**Problem:** The recovery only runs when `admin_config_is_complete()` returns false. If a genesis file provides an admin pubkey, the kind 30078 is never checked, even if it has a different (newer) value.
|
|
**Fix:** Always fetch the kind 30078 `d=agent_config` regardless of whether the genesis already provided values. Compare the fetched admin_pubkey with the loaded one. If they differ, log a `DEBUG_WARN` with both values. Genesis wins (operator intent), but the warning makes the conflict visible in logs.
|
|
|
|
### Fix G: `main()` — always re-publish kind 30078
|
|
**File:** `src/main.c` ~line 1152
|
|
**Problem:** `persist_runtime_config_to_nostr` only runs when `bootstrap_mode || first_run`. On subsequent runs, the kind 30078 is never refreshed. If relays purge the event, it's gone.
|
|
**Fix:** Move `persist_runtime_config_to_nostr(&cfg)` outside the `if (bootstrap_mode || first_run)` block so it runs on every startup. Kind 30078 is a replaceable event — re-publishing is safe, idempotent, and ensures relay persistence. Log the result but don't fail startup if it doesn't succeed.
|
|
|
|
## Execution Order
|
|
|
|
1. Fix E first (relay delivery) — this makes all other wizard publishes more reliable
|
|
2. Fixes A + B (new_agent_flow) — the most critical bug
|
|
3. Fixes C + D (existing_agent_flow) — second most critical
|
|
4. Fix G (always re-publish in main) — ensures long-term relay persistence
|
|
5. Fix F (conflict detection) — nice-to-have logging improvement
|
|
|
|
## Testing
|
|
|
|
- Wizard → New Agent → Install Service: verify kind 30078 exists on relays after service starts
|
|
- Wizard → New Agent → Boot: verify kind 30078 exists on relays
|
|
- Wizard → Existing Agent → Install Service (no changes): verify kind 30078 re-published
|
|
- Wizard → Existing Agent → Boot (no changes): verify kind 30078 re-published
|
|
- `--nsec` only restart: verify kind 30078 recovered and re-published
|
|
- `--config genesis.jsonc` with different admin than kind 30078: verify warning logged, genesis wins, kind 30078 updated
|