AI Suggestions
After every change you make, Specor runs AI agents in the background. Agents analyse the affected part of your product graph and surface suggestions — structured observations about gaps, conflicts, naming issues, and more.
Suggestions appear in the AI Suggestions view (brain icon in the left sidebar).
The inbox
The Inbox tab shows all pending suggestions for the active branch, ordered by severity:
- Error (red) — structural problems that should be fixed
- Warning (amber) — gaps or inconsistencies worth addressing
- Info (blue) — low-priority observations
Each suggestion card shows:
- The agent type that raised it (Coverage, Conflict, Flow, etc.)
- A confidence score — how certain the agent is (shown as a percentage)
- A title and expandable body explaining the issue in plain language
- Which node it refers to
Acting on suggestions
For each suggestion you have three actions:
| Action | What it does |
|---|---|
| AI Fix | Opens the AI Chat panel pre-filled with the suggestion context, ready to apply a fix |
| Snooze | Hides the suggestion until a later date — choose tomorrow, 1 week, 2 weeks, or 1 month |
| Reject | Permanently dismisses the suggestion. Optionally add a note. Tick Don’t suggest again to suppress that category for this node. |
You can also select multiple suggestions with the checkboxes and bulk-snooze or bulk-reject them.
Agent types
Coverage
Checks whether required elements are present for each node type.
- Feature has no linked Metric
- Feature has no Acceptance Criteria
- Feature is not linked to any Need
- Flow has no Actor assigned
- Capability has no child Features
- Metric has no
target_valueset - Need is not addressed by any Feature
- Duplicate feature detected (semantically identical to another)
Conflict
Detects semantic contradictions between connected nodes.
- Two Rules on the same Feature contradict each other
- A Feature’s Rule conflicts with its parent Capability’s constraints
- Steps in a Flow form a cycle (circular dependency)
Flow
Checks the structural integrity of Flows.
- A Flow has no terminal step (no step with no outgoing transitions)
- A Step exists that can’t be reached from the start
- A step transition violates a defined state machine
Impact
Tracks downstream consequences of structural changes.
- A Metric’s Feature was deleted
- A Rule changed on a Feature that has Flows — those Flows may need review
- A Feature was archived but other nodes depend on it
- A Flow has steps but no Actor involvement
Naming
Flags vague or low-quality titles and descriptions.
- Title looks like a placeholder (e.g. “New Feature”, “TODO”)
- Node has no description
- Title is too generic to be useful
- Description is present but gives no real information
Alignment
Checks whether connected nodes are semantically consistent.
- A Feature doesn’t actually address its linked Need
- A Feature is outside its parent Capability’s scope
- A Capability doesn’t fit its Domain
- An Acceptance Criteria tests the wrong thing
- A Metric doesn’t measure what the Feature aims to achieve
Suppressions
If you reject a suggestion with Don’t suggest again, a suppression is created. That agent category will no longer raise suggestions for that specific node.
To review or remove suppressions, click the Suppressions tab in the AI Suggestions view. Deleting a suppression re-enables that check for the node.
Runs log
The Runs tab shows a log of every agent run:
- Which agent ran
- How many suggestions were created, updated, or suppressed
- The timestamp of the run
Use this to understand why a suggestion was or wasn’t raised after a recent change.
Tips
- Use AI Fix liberally. The AI Chat panel launched from a suggestion already has full context — it’s the fastest way to address issues.
- Snooze, don’t ignore. If a suggestion is valid but you can’t address it now, snooze it so it resurfaces at the right time.
- Don’t suppress too broadly. Suppressions are per-node per-category — they’re precise. Use them when a suggestion is intentionally not applicable, not just inconvenient.