Decisions

  • Pending: Search scope — Claude Code repos only, or also similar AI-assistant setups?
  • Pending: Trigger — cron schedule (weekly/monthly) or on-demand only?

User Tasks


Summary

Periodically search GitHub for similar Claude Code repos/setups and extract recommendations for Opus.

Problem / Motivation

Other people build similar systems with Claude Code — custom skills, hooks, agents, vault structures. Without looking externally, we miss patterns, tools, and ideas that could improve Opus. Currently this only happens when the user manually searches.

Proposed Solution

A scheduled task (or skill) that:

  1. Searches GitHub for repos related to Claude Code setups (CLAUDE.md, .claude/ configs, vault structures)
  2. Analyzes interesting findings (structure, skills, hooks, agents, conventions)
  3. Produces a short recommendations report for user review
  4. User decides what to adopt — nothing is auto-applied

Open Questions

No open questions.


Phase Overview

PhaseDescriptionStatus
Phase 1Manual research skill
Phase 2Scheduled cron task

Phase 1: Manual Research Skill —

Goal: A /research-repos skill that searches GitHub and produces a recommendations report.

File / FeatureDetailsOwnerStatus
Define search queriesGitHub search terms for Claude Code setups, CLAUDE.md files, .claude/ configsopus
Build skill.claude/skills/research-repos/SKILL.mdopus
Output formatShort report: what was found, what’s relevant, what to consider adoptingopus

Phase 2: Scheduled Cron —

Goal: Run the research automatically on a schedule and surface findings.

File / FeatureDetailsOwnerStatus
Cron triggerScheduled task (weekly or monthly)opus
Diff from last runOnly surface new/changed repos since last checkopus
NotificationAlert user when new recommendations are availablemv

Test

Manual tests

TestExpectedActualLast
pending-

AI-verified tests

ScenarioExpected behaviorVerification method

E2E tests

ScenarioAssertion

Integration tests

ComponentCoverage

Unit tests

ComponentTestsCoverage

History

DateEventDetails
2026-03-04CreatedUser idea for external benchmarking

References

  • FR-048 (System Self-Review) — internal review; this is external research
  • FR-066 (Self-Improving System) — improvements from internal patterns; this adds external input
  • FR-044 (Task Scheduler) — provides scheduled execution runtime