vibe-learnings

name	vibe-learnings
description	Extract learnings, patterns, and anti-patterns from commit history, session work, and project evolution. Use when: end of a sprint, after a major debug, post-mortem, or to build institutional memory before starting something new. Reads git log + .vibe/ state + produces structured learnings written to context-memory. Wraps: context-memory, knowledge-base skills.
argument-hint	[commits|session|full|retro] [--since YYYY-MM-DD] [--save]
version	1.0.0
allowed-tools	["Read","Write","Bash(git log*)","Bash(git show*)","Bash(git diff*)","Bash(cat .vibe/*)","Bash(ls .vibe/projects/*)"]

Vibe-Learnings — Institutional Memory Extraction

Dispatcher

• No args or commits → extract learnings from git log (last 30 commits by default)
• session → extract from current session's work (recent changes only)
• full → full project retrospective (all commits + all .vibe/ state)
• retro → structured sprint retrospective format (what worked / what didn't / what to change)
• --since {date} → limit to commits after this date
• --save → write learnings to .vibe/context-memory.json automatically

---

Step 1 — Read the Commit History

git log --oneline --all --since="30 days ago"
# Or full history:
git log --oneline --all

For each commit, categorize by what kind of work it represents:

Category	Signal words in message
Feature	feat, add, implement, build, create
Fix	fix, bug, patch, resolve, correct
Revert	revert, remove, undo, rollback, delete
Refactor	refactor, clean, simplify, reorganize
Docs	docs, readme, comment, document
Config	chore, config, deps, package, setup
Experiment	wip, try, test, experiment, attempt

---

Step 2 — Identify Patterns

Analyze the categorized commits to find:

What Worked (Positive Patterns)

Look for: commits that built on each other cleanly, features that shipped without reverts, approaches that became conventions.

Signal: feat → feat → feat sequences with no intervening fix or revert Example output:

✅ PATTERN: {name}
   Evidence: commits {sha1}, {sha2}, {sha3}
   What worked: {description}
   Why it worked: {rationale}
   Repeat when: {conditions}

What Failed (Anti-Patterns)

Look for: reverts, back-to-back fixes, commits that undid previous commits.

Signal: add X followed by remove X or revert X within a few commits Example output:

❌ ANTI-PATTERN: {name}
   Evidence: {sha1} added it, {sha2} removed it
   What went wrong: {description}
   Root cause: {why it happened}
   Avoid when: {conditions to watch for}

Scope Creep Signals

Look for: commits that added things not in the original plan, features that expanded beyond spec.

⚠️  SCOPE CREEP: {what expanded}
   Original intent: {what was planned}
   What was actually built: {what shipped}
   Impact: {time/complexity added}

Technical Debt Accumulation

Look for: // TODO, // FIXME, // HACK patterns in changed files, commits with "temporary" or "workaround" in messages.

---

Step 3 — Architecture Evolution

If 10+ commits exist, trace how the architecture changed:

ARCHITECTURE EVOLUTION:
━━━━━━━━━━━━━━━━━━━━━━
  v0 (initial): {what the structure looked like}
  ↓
  v1 (after {sha}): {what changed and why}
  ↓
  v2 (after {sha}): {what changed and why}
  ↓
  Current: {current structure}

KEY INFLECTION POINTS:
  • {sha}: {why this commit changed the direction}
  • {sha}: {why this commit changed the direction}

---

Step 4 — Structured Retrospective Format (retro mode)

RETROSPECTIVE — {project or sprint name}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
PERIOD: {date range}
COMMITS: {N} | FEATURES: {N} | FIXES: {N} | REVERTS: {N}

WHAT WORKED WELL:
  1. {pattern or decision that paid off}
  2. {tool or approach that saved time}
  3. {process that caught problems early}

WHAT DIDN'T WORK:
  1. {approach that caused rework}
  2. {assumption that was wrong}
  3. {scope that was underestimated}

WHAT TO CHANGE:
  1. {specific change to make next time}
  2. {process improvement}
  3. {technical decision to reconsider}

SURPRISES:
  • {unexpected finding or behavior}

KEY METRICS:
  • Churn rate: {N}% of lines written were later deleted
  • Fix ratio: {N} fixes per {N} features (lower = better design upfront)
  • Revert count: {N} (each revert = a decision made twice)

---

Step 5 — Write to Context Memory (--save)

If --save is passed, append findings to .vibe/context-memory.json:

{
  "learnings": [
    {
      "id": "{timestamp}",
      "type": "pattern|anti-pattern|architecture|retro",
      "title": "{short title}",
      "description": "{full description}",
      "evidence": ["{commit sha}", "{commit sha}"],
      "apply_when": "{conditions}",
      "avoid_when": "{conditions}",
      "source": "vibe-learnings",
      "date": "{ISO 8601}"
    }
  ]
}

These learnings will be available in future sessions via /vibe-status decisions.

---

Output Summary Format

VIBE-LEARNINGS — {scope}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Analyzed: {N} commits | {date range}

✅ PATTERNS ({N} found):
  1. {pattern name} — {one-line summary}
  2. ...

❌ ANTI-PATTERNS ({N} found):
  1. {anti-pattern name} — {one-line summary}
  2. ...

⚠️  WATCH OUT FOR:
  • {risk or recurring issue}

💡 RECOMMENDATIONS FOR NEXT SESSION:
  1. {specific actionable advice}
  2. ...

{If --save}: Written to .vibe/context-memory.json ✅