Computed locally from your own turns in this session. Assistant replies and tool output were dropped before scoring; the raw and scrubbed text were discarded after. Only the signals and recommendations below were kept.
A single readable score, plus the sub-signals behind it.
Session opened with exceptionally clear task goal and good context references, but pivoted to exploratory testing without discussing the scope shift. No major rework loops (disciplined), but scope expanded in mid-flight by adding new tasks rather than staying focused on the original one.
The only thing that would be persisted. No prompt text, no PII — numbers, booleans, and one short tip.
{
"session_type": "other",
"surface": "ide",
"goal_clear_upfront": 0.85,
"done_criteria_upfront": null,
"context_front_loaded": 0.7,
"iteration_tightness": 0.65,
"invited_critique": null,
"had_repro_upfront": false,
"scope_changes": 2,
"late_requirements": 2,
"turns_to_first_artifact": 25,
"rework_turns_est": 0,
"quality_score": 73,
"recommendation": "Discuss upfront whether the session will execute a plan first then test, or explore-then-iterate; mixing both without pre-negotiating scope leads to drift.",
"recommendation_category": "iteration_tightness"
}
Private to you — a couple of gentle ideas, and something you're already doing well.
Clarify whether you will execute the plan first then validate, or validate first then iterate. Mixing both without pre-agreement causes scope to drift.
Before starting: 'I want to execute the plan as-written first, then test' or 'Let me explore behavior first to find gaps.'When exploratory work uncovers new issues, stop to decide if they block the current task or belong in follow-up work, rather than folding them in.
When an issue surfaces: 'Is this a blocker, or should we scope it as a separate task and stay focused on Task 14?'Writing down what counts as done for the current task up-front helps prevent new discoveries from bloating scope.
Upfront: 'Task 14 is complete when [specific criteria]. Everything else goes to the next task.'Correctly scoped duplicate-account handling and signout feature as separate specs rather than patches, and got Opus review before scheduling execution.
Curated by your org for the areas that scored low this session.
25 Effective Prompt Templateshttps://novakit.ai/blog/prompt-engineering-templates-that-workAI Debugging & Error Promptshttps://pub.towardsai.net/the-3-prompt-types-every-sw-engineer-uses-daily-how-to-make-them-better-79891090cc39GitHub: Prompting for Agentic Devshttps://github.com/panaversity/learn-agentic-ai/blob/main/-01_lets_get_started/02_prompts_by_examples/readme.mdWhere clarity was high and where things surfaced a bit later. Your turns; content dropped, only the pattern kept.
| Turn | Pattern | Signal |
|---|---|---|
| 0 | Clear task goal with specific methodology stated upfront. | good |
| 2 | Multiple agents dispatched for parallel execution. | good |
| 25 | Pivoted to manual exploratory testing without discussing scope shift. | drift |
| 71 | Discovered duplicate account behavior; led to new spec creation. | drift |
| 100 | Got Opus review for new spec before scheduling execution. | good |
| 104 | Clean handoff with Task 14 and discovered work prepped for next session. | good |
/pulse-analyse