specstory-yak（幽灵故事牦牛）

Specstory Yak Shave Analyzer

Analyzes your .specstory/history to detect when coding sessions drifted off track from their original goal. Produces a "yak shave score" for each session.

How It Works

1. Parses specstory history files from a date range (or all recent sessions)
2. Extracts the initial user intent from the first message
3. Tracks domain shifts: file references, tool call patterns, goal changes
4. Scores each session from 0 (laser focused) to 100 (maximum yak shave)
5. Summarizes your worst offenders and patterns

What Is Yak Shaving?

"I need to deploy my app, but first I need to fix CI, but first I need to update Node, but first I need to fix my shell config..."

Yak shaving is when you start with Goal A but end up deep in unrelated Task Z. This skill detects that pattern in your AI coding sessions.

Usage

Slash Command

When invoked via /specstory-yak, interpret the user's natural language:

Direct Script Usage

python /path/to/skills/specstory-yak/scripts/analyze.py [options]

Arguments:

• --days N - Analyze last N days (default: 7)
• --from DATE - Start date (YYYY-MM-DD)
• --to DATE - End date (YYYY-MM-DD)
• --path PATH - Path to.specstory/history (auto-detects if not specified)
• --top N - Show top N worst yak shaves (default: 5)
• --json - Output as JSON
• --verbose - Show detailed analysis
• --by-mtime - Filter by file modification time instead of filename date
• -o, --output FILE - Write report to file (auto-adds.md or.json extension)

Examples:

# Analyze last 7 days
python scripts/analyze.py

# Analyze last 30 days, show top 10
python scripts/analyze.py --days 30 --top 10

# Analyze specific date range
python scripts/analyze.py --from 2026-01-01 --to 2026-01-28

# Filter by when files were modified (not session start time)
python scripts/analyze.py --days 7 --by-mtime

# JSON output for further processing
python scripts/analyze.py --days 14 --json

# Save report to a markdown file
python scripts/analyze.py --days 90 -o yak-report.md

# Save JSON to a file
python scripts/analyze.py --days 30 --json -o yak-data.json

Output

Yak Shave Report (2026-01-21 to 2026-01-28)
==========================================

Sessions analyzed: 23
Average yak shave score: 34/100

Top Yak Shaves:
---------------
1. [87/100] "fix button alignment" (2026-01-25)
   Started: CSS fix for button
   Ended up: Rewriting entire build system
   Domain shifts: 4 (ui -> build -> docker -> k8s)

2. [72/100] "add logout feature" (2026-01-23)
   Started: Add logout button
   Ended up: Refactoring auth system + session management
   Domain shifts: 3 (ui -> auth -> database)

3. [65/100] "update readme" (2026-01-22)
   Started: Documentation update
   Ended up: CI pipeline overhaul
   Domain shifts: 2 (docs -> ci -> testing)

Most Focused Sessions:
----------------------
1. [5/100] "explain auth flow" (2026-01-26) - Pure analysis, no drift
2. [8/100] "fix typo in config" (2026-01-24) - Quick surgical fix

Patterns Detected:
------------------
- You yak shave most on: UI tasks (avg 58/100)
- Safest task type: Code review/explanation (avg 12/100)
- Peak yak shave hours: 11pm-2am (avg 71/100)

Scoring Methodology

The yak shave score (0-100) is computed from:

Score interpretation:

• 0-20: Laser focused
• 21-40: Minor tangents
• 41-60: Moderate drift
• 61-80: Significant yak shaving
• 81-100: Epic rabbit hole

Present Results to User

IMPORTANT: After running the analyzer script, you MUST add a personalized LLM-generated summary at the very top of your response, BEFORE showing the raw report output.

LLM Summary Guidelines

Generate a 3-5 sentence personalized commentary that:

1. Opens with a verdict - A witty one-liner about the overall state (e.g., "Your coding sessions this week were... an adventure." or "Remarkably disciplined! Someone's been taking their focus vitamins.")
2. Calls out the highlight - Reference the most notable session specifically:

- If high yak shave: "That January 25th button fix that somehow became a Kubernetes migration? *Chef's kiss* of scope creep." - If low yak shave: "Your January 26th auth flow explanation was surgical - in and out, no detours."

1. Identifies a pattern - Note any recurring theme:

- "You seem to yak shave most when starting with UI tasks" - "Late night sessions are your danger zone" - "Your refactoring sessions tend to stay focused"

1. Ends with actionable advice or a joke - Either:

- A practical tip: "Consider time-boxing those 'quick CSS fixes' - they have a 73% yak shave rate" - Or a joke: "At this rate, your next typo fix will result in a complete rewrite of the Linux kernel"

Example LLM Summary

## 🐃 Your Yak Shave Analysis

Well, well, well. You came to fix buttons and left having rewritten half the
infrastructure. Your average yak shave score of 47/100 puts you firmly in
"classic developer behavior" territory.

The standout? That January 25th session where a CSS alignment fix somehow
evolved into a full Kubernetes deployment overhaul. Four domain shifts later,
you probably forgot what a button even looks like.

Pattern I noticed: Your UI tasks have a 58% higher yak shave rate than your
code review sessions. Maybe start labeling those "quick UI fixes" as
"potential 3-hour adventures" in your calendar.

Here's the full breakdown:

Then show the raw report output below your summary.

What to Highlight

After your summary, when presenting the raw results:

1. The worst offenders with before/after comparison
2. Patterns in when/what causes yak shaving
3. Actionable insight - what task types to watch out for