Why session scope matters
Spur AI works best when each session stays focused on a single topic — one failing suite, one triage question, one authoring task. Two things happen when a session drifts across unrelated topics:- Answer quality drops. Every reply is generated with the full session history in view. Modern LLMs have finite context windows (roughly hundreds of thousands of tokens at the frontier, but effective recall degrades well before that limit — the “lost in the middle” effect). Once earlier messages, screenshots, and tool results start to crowd the window, Spur AI is more likely to conflate topics, miss recent details, or reason from stale context.
- Usage grows fast. Each turn re-processes the entire conversation so far. A session that started with one question and drifted through five more topics will cost significantly more per follow-up than a fresh session asking the same last question in isolation.
When to start a new session
Start a new session — don’t continue the current one — when any of these are true:- You’re switching to a different test, suite, or run than the one you’ve been discussing.
- You’re switching modes — for example, moving from triaging a failure to generating a new test.
- The current session is long (roughly 15–20 back-and-forth turns) and the next question doesn’t depend on that history.
- You’ve attached large context earlier — spreadsheets, PDFs, PRs, Loom recordings — that isn’t relevant to the next question.
- Spur AI starts repeating itself, contradicting earlier answers, or referencing the wrong test. That’s a signal the effective context is saturated.
One session, one topic
Aim for a shape like:- Session A — “Triage the checkout suite failures from last night’s run on staging.”
- Session B — “Generate a new test for the password reset flow from this spec.” (new session, attach the PDF here — not in Session A)
- Session C — “Summarize test health for the mobile suite over the past week.”
Keeping sessions efficient
Within a single session, small habits keep the context tight:- Lead with the specific target. Reference the test, suite, run, or environment by name up front so Spur AI locks onto the right context on turn one instead of hunting for it across several messages.
- Attach only what the current question needs. Every attached spreadsheet, PDF, ticket, PR, or Loom sits in the session context from the moment you add it. Attach them in the session that needs them, not preemptively.
- Ask one thing at a time. Multi-part questions (“summarize the run, then generate three new tests, then open a Jira ticket”) produce longer answers and longer follow-ups. Break them into separate turns, or separate sessions, and the whole thing runs cheaper and more accurately.
- Wrap up before pivoting. If you’re about to change topics, ask Spur AI to summarize the outcome of the current thread first, then start a new session and paste the summary in if you need to carry anything forward. This is far leaner than dragging the entire history into the next topic.
- Don’t paste huge logs or transcripts inline. Reference the run or attach the file — inline blobs sit in the context permanently and count against every subsequent turn.
Signs it’s time to start over
Open a new session when you notice:- Replies are getting slower or shorter than earlier in the same session.
- Spur AI is answering the previous question instead of the current one.
- You’re correcting the same misunderstanding more than once.
- The todo panel is carrying tasks from an unrelated earlier request.