analyse-team-session_skill

Overview

This skill analyses an agent team session export against the official Claude Code agent teams best practices and produces a structured, actionable report. It returns a suitability verdict, an 8-category scorecard, evidence-backed explanations, targeted recommendations, and a ready-to-use improved prompt rewrite. Use it when you need concrete guidance on whether an agent team run was effective and how to fix specific problems.

How this skill works

The skill ingests a Markdown team session transcript, extracts session metadata (team name, session ID, duration, agents, original user prompt, task list, messages, tool calls, and final output), fetches the official agent teams best practices, and evaluates the session across eight rubric categories. For each category it rates performance (Strong/Partial/Gap), quotes session evidence, and cites the corresponding best-practice passage. It then generates a prioritized recommendation list and a full improved prompt ready to copy-paste.

When to use it

• You have a team session export and want a structured, evidence-backed review.
• You ran an agent team and need to know what went wrong or how to improve next time.
• You want to confirm whether a task was a good fit for agent teams or should be done in a single session.
• You need a ready-to-run improved spawn prompt that follows official best practices.
• You want a precise checklist for cleanup, cost control, and lead orchestration after a run.

Best practices

• Always fetch and reference the official agent teams docs before scoring; assessments must link session evidence to specific doc guidance.
• Extract and quote specific session passages for every rating—no vague statements.
• Evaluate all eight rubric categories: Suitability, Context Sharing, Task Sizing, Communication Quality, File Conflict Avoidance, Lead Orchestration, Cost Efficiency, Cleanup.
• Produce 3–5 prioritized, actionable recommendations that map directly to observed evidence and a cited doc rule.
• Provide a complete improved prompt that the user can copy-paste, incorporating spawn prompt context, task scope, deliverable format, and cleanup instructions.

Example use cases

• A developer submits a team session export and asks, 'What went wrong with this run?' and needs a concrete fix plan.
• A product manager wants to know if a research task was a good fit for agent teams or should be refactored.
• A team lead needs cost and cleanup recommendations after an expensive, long-running team.
• You want an improved, compliant spawn prompt to re-run the same job with better context and orchestration.
• A reviewer needs a scorecard showing which best practices were followed and exact lines from the transcript as evidence.