Automation: sources, rules, and judgment
Automation is MateBot’s watching layer. It is off by default, and even when on it starts with a processing mode.
Processing modes
| Mode | Behavior |
|---|---|
| Off | No source is checked and nothing is recorded |
| Record only | MateBot still judges, but only records what it would have handled; no action is taken. Start here |
| Auto-create | On a rule match, MateBot opens a board card in the rule’s target project; rules without a target still only record |
Auto-create has its own switch and a Pause all auto-create control. Pausing keeps every setting and keeps recording.
Sources
Sources come from plugins or connectors. Nothing is watched until you enable it:
- DMs: what people send you directly
- Groups: messages that @mention the channel bot in a group (MateBot itself is not in the group); bodies kept for 30 days; only the starter can continue the thread; off by default
- GitHub: PRs awaiting your review and issues/PRs assigned to you, ready after sign-in
- Connector polling: a generic connector reads new content by field mapping and tracks its position with a cursor; fits tickets, forms, and internal systems
- Other channels: appear in the source list after you install the channel or plugin
Each source has its own check interval (1 / 5 / 15 minutes) plus last sync, last error, and health. Incomplete configuration names the missing fields instead of polling blindly.
Rules
A rule answers two questions: what counts as a match, and what happens then.
- Built-in rules: bug report, blocked or waiting, directly assigned to the assistant
- Custom rules: say what counts, what does not, and what to do when unsure; you can attach your original wording
- On match: record only, or open a board card in the target project
- Auto-create limits: daily cap, cooldown, and observation period; failures are counted and repeated failures pause the rule with a reason
- Rule cap: there is a limit on enabled rules; multiple matches on one message are handled once
Judgment
MateBot runs assessment in day-bucketed observation sessions kept for 30 days. Judgment includes open items and items closed in the last 24 hours, so a follow-up is not mistaken for new work.
The Assessment panel on the right shows each message’s verdict: which rule matched, whether it needs handling, and a failure code when it does not. A record can be turned into an item or dismissed in one click.
DM bridging
DM bridging is enabled per bound channel and is off by default. Once enabled, the owner’s DMs go straight into MateBot’s session and replies return to the same DM. Groups are never bridged.
If a DM maps to exactly one open item and that item is bound to an executor, the message goes straight to the executor without a second retelling.
A sensible rollout
- Start with Record only, connect one source, and let it run for two or three days.
- Read the assessment panel and write the misses into the rule’s “does not count” list.
- When the rule is stable, give it a target project and switch to Auto-create.
- Start with a daily cap of 1–2 cards and widen it after the observation period.
Failures and health
A failed source read is labeled Read failed; it does not mean you have no sources. Aggressive intervals consume quota. When a connector’s fields change, read Last error before editing the field mapping.