メニュー
Cross-session analytics — parse JSONL session files to detect patterns and friction hotspots
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Automates browser interactions for web testing, form filling, screenshots, and data extraction. Use when the user needs to navigate websites, interact with web pages, fill forms, take screenshots, test web applications, or extract information from web pages.
Save current session state to a handover document for seamless continuation in a new session. Invoke manually or triggered automatically when context window reaches 85%.
Use when the user wants to delegate a task to OpenAI Codex CLI, get a second opinion, or run codex review on changes
Configurable 9-phase feature development workflow enforcer (Phase 0 Context Setup → Phase 8 Approval & Commit). Delegates planning to OpenSpec (proposal→specs→design→tasks) and enhances execution with superpowers skills (TDD, subagent-driven-development, worktrees, verification). Auto-validates phase transitions, manages state via context.md, and orchestrates specialized agents. Use when user says "start feature" or "resume feature", mentions implementing/building/creating a feature, needs structured development process, references TDD or test-driven development, asks about feature planning or implementation phases, references 9-phase workflow or mentions "Phase 1" or "Planning phase", needs approval gates or phase validation, or asks "how do I start a feature" or "what's the development workflow".
Structured E2E test troubleshooting using the 6-layer diagnostic ladder (Frontend → API → SP/Schema → Infra → CI/CD). Use when any E2E test fails, UI doesn't render, auth drops, unexpected status codes, CI passes locally but fails in pipeline, or any test failure where the root cause is unclear. Routes to the correct read-only diagnostic agent based on the symptom, walks the sequential ladder with [EVIDENCE]/[HANDOFF]/[ROOT-CAUSE] taxonomy, and enforces the Fix Gate (no action before root cause confirmed).
Post-session retrospective — append tagged learnings (with severity) to configured learnings file (default: `learnings.md`)
| name | project-insights |
| description | Cross-session analytics — parse JSONL session files to detect patterns and friction hotspots |
Analyze Claude Code JSONL session files to identify recurring patterns, friction hotspots, and generate evidence-based CLAUDE.md recommendations.
/project-insights # Default: last 30 days, top 10
/project-insights --days 7 # Last 7 days only
/project-insights --top 5 # Top 5 results per category
Claude Code stores session files at ~/.claude/projects/<mangled-path>/ where the mangled path replaces / with -.
Compute the sessions directory dynamically:
PROJECT_DIR=$(pwd)
MANGLED=$(echo "$PROJECT_DIR" | sed 's|/|-|g')
SESSIONS_DIR="$HOME/.claude/projects/$MANGLED"
Verify the directory exists before proceeding. If not found, ask the user.
Execute the Python extraction script:
python3 ~/.claude/skills/project-insights/scripts/extract.py \
--sessions-dir "$SESSIONS_DIR" \
--days {DAYS} --top {TOP}
Pass through any --days or --top arguments from the user's command. Default: --days 30 --top 10.
Timeout: 5 minutes max. If the script fails, report the error and suggest reducing --days.
The script outputs a JSON summary to stdout. Parse it and present a formatted report.
Format the JSON output into a readable report with these sections:
After successful analysis, update the stamp file:
date +%s > /tmp/claude-insights-last-run.txt
This is used by the insights-reminder hook to determine when to suggest re-running.