weekly-review-builder（每周评论生成器）

Weekly Review Builder

Compress one week's activity into an updated, honest project state. Prevent long-term drift.

Input

Required:

• project_card — full project card with current state, goals, phase
• weekly_daily_logs — array of 5-7 daily loop logs from this week
• weekly_decisions — decisions made during the week
• weekly_outputs — tangible outputs produced this week
• stale_tasks — tasks that were planned but not completed

Optional:

• blocked_tasks — tasks that failed due to blockers

Output Schema

updated_current_phase: string       # refreshed project phase
updated_current_bottleneck: string  # the one most critical blocker now
next_week_goal: string              # single clear goal for next week
next_3_actions: string[]            # top 3 actions for next week
risk_check: string                  # honest assessment of current risks
one_page_project_summary: string    # compressed project status (≤300 words)
review_confidence: "high" | "medium" | "low"
review_writeback: object           # structured record for project memory
stale_tasks_resolved: string[]      # tasks cleared this week
stale_tasks_remaining: string[]     # tasks still pending

Rules

1. Must remove outdated actions. If a planned task wasn't touched, mark it stale and either drop or reschedule it.
2. Must identify the single current bottleneck. Not 3, not 5 — one.
3. Must distinguish progress from activity. Meetings attended ≠ progress. Code shipped ≠ impact. Be precise.
4. Must be honest if no real progress was made. Say so. A stagnant review is better than a flattering one.
5. Evidence-based only. Do not update phase or bottleneck without supporting evidence from logs.

Review Phases

Phase 1: Log Compression

• Read all weekly_daily_logs
• Extract: findings, decisions, completions, blockers
• Separate actual progress from busy-work

Phase 2: Gap Analysis

• Compare weekly_outputs against weekly_daily_logs stated objectives
• Identify: what was claimed vs what was delivered
• Mark stale_tasks as resolved or remaining

Phase 3: State Refresh

• Update current_phase based on evidence (not ambition)
• Identify current bottleneck (the one thing blocking most progress)
• Assess risk: what's the biggest threat to the project right now?

Phase 4: Forward Planning

• Set next_week_goal (one clear goal)
• List next_3_actions (ranked by impact)
• Write one_page_project_summary for human readability

Phase 5: Writeback

• Populate review_writeback with structured record
• Set review_confidence based on evidence quality

Failure Handling

If evidence is insufficient (e.g., logs are missing, can't determine progress):

• Mark review_confidence = "low"
• Do not fabricate phase changes or bottleneck updates
• Request missing log recovery before completing review
• Write partial review with explicit gaps noted