insight-harness

Files (4)

• README.md
• SKILL.md
• hero.png · assets/
• extract.py · scripts/

SKILL.md (first lines)

# Insight Harness

Generate a comprehensive profile of your Claude Code harness configuration and usage patterns over the last 30 days. This is a **superset of /insights** — it includes everything /insights provides, plus token usage, tool breakdowns, skill inventory, hooks, agent patterns, and more.

## What This Does

Runs a Python extraction script that reads harness metadata from:

- `~/.claude/settings.json` (hooks, plugins, permissions)
- `~/.claude/plugins/installed_plugins.json` (plugin inventory)
- `~/.claude/skills/` (skill frontmatter — names and allowed-tools only)
- `~/.claude/usage-data/session-meta/` (pre-computed session stats)
- `~/.claude/projects/*/*.jsonl` (field-whitelisted: tool names, skill names, hook events, tool transition sequences, workflow phase classifications only)
- `~/.claude/projects/*/settings.local.json` (approved permissions)

**Privacy guarantee:** The script uses a strict field whitelist. It NEVER reads tool arguments, message text, tool results, file paths, or any project-specific content. Real credentials exist in JSONL files — the script never touches those fields.

## What You Get (vs /insights)

Everything from /insights, plus:

- **Token usage** — input, output, and total token consumption
- **Tool usage breakdown** — Read, Edit, Bash, Grep, etc. with counts
- **Skills & plugins inventory** — what's installed, invocation frequency
- **Hooks** — configured hooks and how often they fire
- **CLI commands** — which bash commands you run most
- **Agent dispatch patterns** — subagent types, models, background usage
- **File operation style** — Edit vs Write ratio
- **Models used** — Opus, Sonnet, Haiku distribution
- **Permission modes** — how you configure access
- **MCP servers** — connected servers
- **Workflow phases** — classifies tool usage into phases (exploration, implementation, testing, shipping, orchestration) and shows the distribution across sessions
- **Phase transitions** — tracks how you move between workflow phases (e.g., exploration -> implementation), with statistics on disciplined patterns like "test before ship"
- **Tool transitions** — tracks sequential tool usage patterns within turns (e.g., Read -> Edit), showing your most common tool flows

## How to Run

Run the extraction script and open the result:

```bash
python3 ~/.claude/skills/insight-harness/scripts/extract.py
```

The script outputs the HTML file path to stdout. Open it in the browser:

```bash
open "$(python3 ~/.claude/skills/insight-harness/scripts/extract.py)"
```

## Updating

The skill checks for updates automatically when you run it. To update manually:

```bash
python3 ~/.claude/skills/insight-harness/scripts/extract.py --update
```

Or re-run the install command:

```bash
curl -sL https://github.com/craigdossantos/claude-toolkit/archive/main.tar.gz | tar xz -C /tmp && cp -r /tmp/claude-toolkit-main/skills/insight-harness ~/.claude/skills/ && rm -rf /tmp/claude-toolkit-main
```

After running, tell the user:

1. Where the report was saved
2. That it's been opened in their browser
3. A brief summary of the top-level stats (sessions, tokens, skills used, hooks active)
4. They can upload the report to insightharness.com to share their profile publicly

From kabirdos/claude-toolkit, MIT.

More from kabirdos/claude-toolkit