Diary

Diary

Capture insights from your Claude Code sessions — what was decided, what was built, what patterns emerged. Designed to build a searchable record of your work over time.

Connectors

No external connectors needed. Uses Claude Code built-in tools (Read, Write, Bash).

Setup

This skill reads your Claude Code session history and writes diary entries to a folder you choose. A little one-time setup is needed:

1. Create your diary folder

Pick a location where diary entries will live. This should be a folder you can search later — ideally version-controlled so entries are backed up.

mkdir -p ~/diary
cd ~/diary && git init   # optional but recommended

Update the Diary entries path in env.md to match.

2. Find your session transcripts

Claude Code stores session data locally. The path depends on which project you're working in:

~/.claude/projects/[project-path-with-dashes]/sessions-index.json

For example, if your project is at /Users/you/code/my-app, the sessions index would be at:

~/.claude/projects/-Users-you-code-my-app/sessions-index.json

To find yours, run:

ls ~/.claude/projects/

Look for the directory that matches your project path (with slashes replaced by dashes). Update the Session transcripts path in env.md.

3. Confirm env.md paths

Your env.md should have:

Setting	Path
Diary entries	~/diary/ (or your chosen location)
Session transcripts	~/.claude/projects/-Your-project-path/

Workflow

Step 1: Find Sessions to Process

1. Get cutoff date from the most recent diary entry:

   ls ~/diary/*.md | sort | tail -1
   

   Extract the date from the filename (YYYY-MM-DD pattern).

2. Read the sessions index (path from env.md):

   ~/.claude/projects/[your-project-path]/sessions-index.json
   

3. Filter sessions to process — include sessions where:

   • Modified after the cutoff date
   • Has 6+ messages (skip trivial sessions)
   • Is not a side-chain
   • Has a summary or first prompt to reference

Step 2: Analyze Sessions

For each session, extract:

• Key decisions — What was decided and why
• What was built — Features, fixes, changes
• Patterns observed — Recurring themes, approaches that worked or didn't
• Tool efficiency (optional) — Were tools used well? Repeated reads? Sequential calls that could be parallel?

Step 3: Present Summary

Show a table for review before writing:

| Date | Summary | Messages | Efficiency |
|------|---------|----------|------------|
| Mar 1 | Built email skill with contact lookup | 42 | good |
| Mar 1 | Debugged calendar auth flow | 28 | mixed |
| Feb 28 | Research on MCP server patterns | 65 | good |

Sessions found: X sessions since [cutoff date]

If no sessions found: "All caught up! No new sessions since [date]."

Step 4: Generate Entry

Create a consolidated entry covering all sessions:

Filename: YYYY-MM-DD-catchup.md

---
date: YYYY-MM-DD
tags: [tag1, tag2, tag3]
sessions: N
---

## [Date Range or Theme]

### [Date]

**[Summary 1]**
- Key decisions or insights
- What was accomplished

**[Summary 2]**
- Key decisions or insights

### Themes
- [Recurring patterns across sessions]
- [Technical focus areas]

Step 5: Save (and optionally push)

Write to your diary folder, then optionally commit and push:

cd ~/diary && git add . && git commit -m "Add diary: [date range summary]" && git push

Error Handling

Scenario	Response
No new sessions	"All caught up! No sessions since [date]."
Sessions without summaries	Use the first prompt (first 100 chars) as fallback
Very long gap since last entry	Group by week, summarize themes

Customization Notes

How to adapt this skill to your own setup.

• Tags — The tag categories in the entry format are a starting point. Add domain-specific tags for your work (e.g., frontend, data-pipeline, client-work).
• Tool efficiency tracking — This is optional but valuable over time. You can add a script that parses session transcript JSONL files and reports: total tool calls, breakdown by tool, files read 3+ times (inefficiency signal), and sequential calls that could be parallel. Track a simple score (good / mixed / needs improvement) per session. Over time, /reflect can spot patterns like "you keep re-reading the same config file" and propose workflow improvements.
• Entry format — Adjust the markdown template to match what you find useful to look back on. Some people prefer more narrative; others want just bullet points.
• Current session mode — Add a /diary this variant that captures just the current conversation without scanning past sessions.
• Integration with reflect — After 3-5 diary entries, the reflect skill can analyze patterns and suggest system improvements.