What task context is
The task context is the sentence you give the classifier that defines what counts as on-task. The model reads your screenshot AND your task context and asks: does this content match the stated intent? A vague task means a vague classifier; a specific task means precise interventions.
Bad → better → best
- Bad: "work" — the model has no idea what work means for you. Wikipedia, Slack, and YouTube tutorials all plausibly fit.
- Better: "writing my Q2 strategy doc" — the model knows you're writing, can recognize writing tools, and can flag obvious distractions.
- Best: "writing my Q2 strategy doc in Notion; specifically the section on market sizing" — the model can now tell whether you're in Notion, on the right page, and whether you've drifted to tangentially-related research.
Why specificity helps both directions
More specific contexts reduce false positives (the model knows market-research tabs are on-task for a market-sizing doc) AND false negatives (the model knows scrolling Twitter is off-task even if the tweet is about strategy). Vagueness pushes both error rates up.
How to set it
- Open the block editor for an AI-powered block.
- Fill the Task Context field. Limit: 500 characters.
- Save. The new context applies on the next classifier poll.
Update the task context throughout the day as your focus shifts. "Drafting the email" at 10am and "reviewing the contract" at 2pm are different jobs — keep the context current.
Empty task context disables the classifier for that block. The block falls back to pure rule-based blocking against its items list.