Weekly report
🔮 Claude Code antipattern detector & auto-patcher — learns from session logs, patches CLAUDE.md, tracks rule effectiveness
npx -y skills add cmg8431/claude-dejavu --skill weekly-reportAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Generate a weekly report of detected patterns and rule effectiveness. Use when user asks "dejavu report", "what did you learn this week", "dejavu weekly".
SKILL.md
1.8 KB, as published. Nobody here has run it
claude-dejavu: Weekly Report
You are generating a weekly effectiveness report for claude-dejavu.
Steps
- First, gather statistics:
claude-dejavu stats --path "$(pwd)"
- Then list all rules with their grades:
claude-dejavu list --path "$(pwd)"
- Format a report with the following sections:
New Patterns This Week
List any rules created in the last 7 days. If stats show no new rules, say "No new patterns detected this week."
Rules That Fired (Prevented Mistakes)
Highlight rules with fire_count > 0 that have fired recently. These are the rules that actively prevented Claude from repeating past mistakes. Show each rule's grade badge and fire count.
Dead Rules to Consider Removing
List rules with status "dead" or rules that have not fired in 14+ days with low confidence (< 0.6). Suggest the user remove these to keep the rule set lean.
Effectiveness Trend
Summarize overall health:
- Total active rules
- Average confidence across active rules
- Total fires this period
- Grade distribution (how many A/B/C/D rules)
Present the report in a clean, readable format:
=== claude-dejavu Weekly Report ===
New patterns: 2 discovered
Rules fired: 5 times (prevented 5 mistakes)
Dead rules: 1 candidate for removal
Overall health: Good (avg confidence: 0.82)
Grade distribution: A:3 B:2 C:1 D:0
Important
- Be concise but informative
- Highlight wins (rules that fired = mistakes prevented)
- If there are no rules yet, encourage the user to run more sessions so dejavu can learn
- If effectiveness is declining (many dead rules, low confidence), suggest running a fresh scan