DeepSeek Harness (DSH) users accumulate large amounts of conversation and tool-call data over long-term use. Reviewing this data intuitively and understanding the behavior patterns of AI pair programming often lacks sufficient tooling support. dsh-dev-wrapped aims to address this issue by analyzing DSH session data and generating visualized annual report cards.

What is this

This is a plugin for generating DeepSeek Harness (DSH) developer annual reports. It adopts a Spotify Wrapped-style programming retrospective format, analyzes users’ behavior in AI pair programming, and generates a single-file HTML report card for sharing. The plugin is maintained by SleepEggTart, and its core value lies in providing “100% local parsing” and “honest statistics”.

Core features

The plugin mainly provides the following capabilities:

  • Data parsing and statistics: Supports scanning ~/.dsh/sessions (zstd-compressed JSONL) data. It prefers invoking the system zstd CLI for streaming decompression and automatically falls back to fzstd if zstd is not installed.
  • Adapter architecture: Decouples the parsing layer from the statistics layer through the NormalizedEvent model. Supports multiple data sources such as DSH and Claude Code, and supports --adapter all for merging across data sources.
  • Report modes: Provides two modes. The default is a screen-by-screen narrative report (scroll-snap); use --compact to switch to a compact single-page card.
  • In-depth statistics: Includes a tool error-rate ranking, the proportion of late-night coding (0-6), weekday/weekend distribution, model distribution, and DSH agentPreset distribution.
  • Achievement badges: Includes 11 built-in gamified badges (such as Night-Shift Coder and Weekend Warrior), with quantified bronze/silver/gold tiers; it also has 2 hidden easter-egg badges (such as 996 Warning).
  • Year-over-year comparison: Uses --compare to compare this year and last year by sessions, turns, tool calls, active days, and Tokens, showing the growth curve.
  • Month-by-month replay: Reviews months on separate screens, rendering only months with activity.
  • AI summary prompt: Each run outputs *-prompt.txt; copy and paste it back into DSH or Claude Code to generate a zero-cost, local, personalized summary.
  • Developer persona: Generates six persona profiles (such as Midnight Architect and Morning Light Commander) based on work rhythm (peak hours) and style (tool-call volume).
  • Engineering completeness: The project includes CI (Node 18/20/22 matrix), version management (changesets), and fuzz fault-tolerance testing.

Installation and enablement

Install via DSH plugin management (recommended):

dsh plugin --profile <你的profile名> add dsh-dev-wrapped

Install directly from GitHub (alternative):

dsh plugin --profile <你的profile名> add github:SleepEggTart/dsh-dev-wrapped

After installation, enter a slash command in the DSH dialog box to generate a report:

/wrapped                    # 生成报告（默认 DSH 数据源）
/wrapped --adapter all      # 合并 DSH + Claude Code

Typical usage

The following are common invocation methods:

  1. Default DSH data source:
    /wrapped
  1. Cross-data-source merge:
    /wrapped --adapter all
  1. Year-over-year comparison and compact mode:
    /wrapped --year 2026 --compare --compact
  1. CLI local run (specify date range):
    dsh-dev-wrapped --since 2026-08-01 --until 2026-08-31 --include-subagents
  1. Scan Claude Code sessions:
    npx dsh-dev-wrapped --adapter claude-code
  1. Automatically detect data source:
    dsh-dev-wrapped --adapter auto
  1. Annual review + cost estimation + English card:
    dsh-dev-wrapped --year 2026 --estimate-cost --lang en

Applicable scenarios and notes

It is suitable for developers who need to review the efficiency of working with AI and analyze coding habits.

Before use, pay attention to the following:

  1. Data format changes: DSH is in the developer preview stage, and the ~/.dsh/sessions data format may undergo breaking changes as versions evolve.
  2. Runtime permissions: The plugin runs with the permissions of the current DSH process; review the source code and license before installation.
  3. Data integrity: Session files currently being written (while DSH is running) may contain incomplete data; corrupted files are skipped with a warning.
  4. Platform dependencies: On Windows, zstd must be available in PATH; if it is missing, the tool automatically falls back to fzstd; if both are unavailable, the CLI outputs installation commands for the relevant platform.

Summary

dsh-dev-wrapped provides a complete engineering-oriented solution that transforms scattered DSH session data into a structured annual report. Whether your goal is to quantify AI tool usage efficiency or to find hidden developer personas and achievement badges, everything can be completed locally without relying on any cloud service.