跳转至

Use Cases

Real-world scenarios showing how zhi CLI fits into your workflow.

Table of Contents

I. Daily Office

1.1 Summarize Meeting Minutes

Scenario: After a product review meeting, you have a lengthy transcript and need to produce a concise summary to share with the team.

Steps:

Use the built-in summarize skill:

$ zhi run summarize product-review-notes.txt

zhi reads the file and generates a structured summary, saving it to zhi-output/:

[file_read] Reading: product-review-notes.txt (12.3KB)
[file_write] Writing: product-review-notes-summary.md (1.8KB)

File saved: zhi-output/product-review-notes-summary.md

Result: A Markdown summary containing key decisions, action items, and owners. Typically 10-20% of the original length.

For more specific instructions, use interactive mode:

$ zhi
You [approve]: Summarize this meeting transcript, focusing on action items and deadlines

1.2 Translate Documents

Scenario: You received a technical specification in English and need a Chinese translation for team review.

Steps:

Use the built-in translate skill:

$ zhi run translate api-specification.md

The default target language is Chinese. For other directions:

$ zhi -c "Translate project-intro.md to English, keeping Markdown formatting intact"

Output:

[file_read] Reading: api-specification.md (8.5KB)
[file_write] Writing: api-specification-translated.md (9.2KB)

File saved: zhi-output/api-specification-translated.md

Result: A complete translation that preserves document structure (headings, code blocks, lists). Technical terms like API, JSON, and HTTP are kept in English.

1.3 Generate Weekly/Daily Reports

Scenario: Friday afternoon, you need a weekly report based on your work notes and Git history.

Steps:

$ zhi
You [approve]: Generate a weekly report from my work notes, including completed items, \
...  in-progress items, and plans for next week

[file_read] Reading: weekly-notes.md (3.2KB)
  → Allow reading this file? (y/n) y
[file_write] Writing: weekly-report-2026-02-23.md (1.5KB)
  → Allow creating this file? (y/n) y

File saved: zhi-output/weekly-report-2026-02-23.md

Result: A formatted report organized into "Completed", "In Progress", and "Next Week" sections.

Combine with Git history using pipes:

$ git log --oneline --since="last monday" | zhi -c "Generate a work summary based on these Git commits"

II. Data Extraction & Comparison

2.1 Compare Two Document Versions

Scenario: A client sent a revised contract. You need to quickly identify all changes between versions.

Steps:

Use the built-in compare skill:

$ zhi run compare contract-v1.md contract-v2.md

zhi compares both documents paragraph by paragraph:

[file_read] Reading: contract-v1.md (6.2KB)
[file_read] Reading: contract-v2.md (6.8KB)
[file_write] Writing: contract-diff-report.md (2.5KB)

File saved: zhi-output/contract-diff-report.md

Result: A structured diff report showing added, removed, and modified content with a comparison table highlighting specific changes and their locations.

2.2 Extract Key Figures from Financial Reports

Scenario: You received a quarterly financial report as a PDF and need to extract key metrics for management.

Steps:

$ zhi -c "Read quarterly-report.pdf, extract revenue, profit, and YoY growth figures, save as a spreadsheet"

[ocr] Recognizing: quarterly-report.pdf (4.5MB)
[file_write] Writing: financial-metrics.xlsx (3.2KB)

File saved: zhi-output/financial-metrics.xlsx

Result: Key financial data extracted from the PDF and organized into a structured table with calculated growth percentages.

2.3 Clean Up a Contact List

Scenario: You have a messy CSV of client contacts with duplicates and inconsistent formatting.

Steps:

$ zhi -c "Read contacts.csv, remove duplicates, sort by company name, save as a new Excel file"

[file_read] Reading: contacts.csv (8.6KB)
[file_write] Writing: contacts-cleaned.xlsx (5.1KB)

File saved: zhi-output/contacts-cleaned.xlsx

Result: Duplicates removed (matched by name + phone), formatting unified, and records sorted alphabetically by company name. The original CSV is untouched.

2.4 Cross-Document Summary

Scenario: You have multiple project progress reports and need a consolidated overview.

Steps:

$ zhi
You [approve]: Read project-A-status.md, project-B-status.md, and project-C-status.md. \
...  Extract completion progress, risks, and next steps from each, then produce a summary report

[file_read] Reading: project-A-status.md (3.5KB)
[file_read] Reading: project-B-status.md (4.1KB)
[file_read] Reading: project-C-status.md (2.8KB)
[file_write] Writing: project-summary.md (3.6KB)

File saved: zhi-output/project-summary.md

Result: A consolidated report with a progress overview table, risk items, and action plans extracted from all three reports.

III. Content Creation

3.1 Article Summary and Rewrite

Scenario: You need to condense a long article or rewrite it for a different audience.

Steps:

Generate a summary:

$ zhi run summarize tech-article.md

Rewrite for a different audience:

$ zhi
You [approve]: Read tech-article.md and rewrite it in plain language for a non-technical audience

[file_read] Reading: tech-article.md (6.8KB)
[file_write] Writing: tech-article-simplified.md (5.2KB)

File saved: zhi-output/tech-article-simplified.md

Result: A rewritten version with adjusted language and technical depth, preserving the core message. The original file is not modified.

3.2 Multilingual Content Translation

Scenario: Your product needs bilingual release notes published simultaneously.

Steps:

$ zhi -c "Read changelog-v2.5.md, translate to English, keep Markdown formatting and version numbers"

[file_read] Reading: changelog-v2.5.md (2.1KB)
[file_write] Writing: changelog-v2.5-en.md (2.3KB)

File saved: zhi-output/changelog-v2.5-en.md

Result: An accurate, fluent translation with formatting fully preserved. Version numbers, code snippets, and variable names are left untranslated.

3.3 Markdown Formatting

Scenario: You have raw, unstructured notes that need to be formatted into proper Markdown.

Steps:

$ zhi
You [approve]: Read meeting-notes-raw.txt, organize into well-formatted Markdown with headings and lists

[file_read] Reading: meeting-notes-raw.txt (1.8KB)
[file_write] Writing: meeting-notes-formatted.md (2.0KB)

File saved: zhi-output/meeting-notes-formatted.md

Result: A clean Markdown file with proper heading hierarchy, ordered/unordered lists, and block quotes. Ready to paste into Notion, GitHub, or any Markdown-compatible platform.

IV. Data Processing

4.1 PDF/Image OCR Text Extraction

Scenario: You have a scanned contract PDF or a photo of a whiteboard that needs to be converted to editable text.

Steps:

Extract text from a PDF:

$ zhi -c "Use OCR to extract text from contract-scan.pdf, save as a text file"

[ocr] Recognizing: contract-scan.pdf (3.2MB)
[file_write] Writing: contract-content.txt (8.5KB)

File saved: zhi-output/contract-content.txt

Extract text from an image:

$ zhi -c "Extract text from whiteboard-photo.jpg, organize into a bulleted list"

Result: Text extracted from PDF or image files, cleaned up, and formatted as requested. Supports PDF, PNG, JPG, GIF, WEBP. Maximum file size: 20MB.

Batch processing:

$ zhi
You [approve]: Extract text from screenshot1.png and screenshot2.png, combine into a single document

4.2 CSV/Excel Data Cleanup

Scenario: You have a messy CSV that needs cleaning, aggregation, and conversion to Excel.

Steps:

$ zhi
You [approve]: Read sales-data-raw.csv, aggregate sales by region, generate a new Excel file

[file_read] Reading: sales-data-raw.csv (15.6KB)
[file_write] Writing: sales-by-region.xlsx (4.8KB)

File saved: zhi-output/sales-by-region.xlsx

CSV-to-CSV processing also works:

$ zhi -c "Read user-list.csv, remove duplicates, sort by registration date, save as a new CSV"

Result: Data cleaned, aggregated, or transformed and saved in the requested format. Excel (.xlsx) and Word (.docx) output are supported out of the box.

4.3 Web Content Scraping and Analysis

Scenario: You need to pull information from a web page and organize it into a report.

Steps:

$ zhi
You [approve]: Fetch https://example.com/tech-trends-2026, extract key points, and save as a summary

[web_fetch] Fetching: https://example.com/tech-trends-2026 (28.5KB)
[file_write] Writing: tech-trends-summary.md (2.1KB)

File saved: zhi-output/tech-trends-summary.md

Result: Main content extracted from the web page (ads and navigation stripped), organized into a structured summary. Web content is capped at 50KB.

Combine with translation:

$ zhi -c "Fetch https://example.com/release-notes, translate to Chinese, and save"

V. Custom Skills

zhi lets you define reusable skills as YAML files. Once created, they work just like built-in features. Skills run on glm-4-flash, costing roughly 10% of interactive chat.

5.1 Contract Review Skill

Scenario: Your legal team needs a standardized checklist applied to every new contract.

Steps:

Let zhi create the skill:

$ zhi
You [approve]: Create a contract review skill that checks for: key clauses, \
...  deadlines, liability limits, termination conditions, and payment terms

[skill_create] Creating skill: contract-review
  → Allow creating this skill? (y/n) y

Skill 'contract-review' created

Use the skill:

$ zhi run contract-review vendor-agreement.md

Result: A structured review report with pass/fail/attention status for each checklist item, plus specific recommendations.

Tip

zhi also ships with a built-in contract-review composite skill that chains analyze, compare, and proofread together. Use the built-in one for comprehensive reviews, or create a custom one tailored to your specific checklist.

5.2 Email Draft Skill

Scenario: You send multiple similar business emails daily and want to automate draft generation.

Steps:

$ zhi
You [approve]: Create an email draft skill that generates formal business emails from brief descriptions

[skill_create] Creating skill: email-draft
  → Allow creating this skill? (y/n) y

Skill 'email-draft' created

Use it:

$ zhi -c "Use the email-draft skill: notify client Wang about the project acceptance being postponed to next Wednesday"

Result: A properly formatted business email draft ready to copy into your email client.

5.3 Data Analysis Skill

Scenario: Every month you need to produce a sales analysis report in a fixed format.

Steps:

$ zhi
You [approve]: Create a monthly data analysis skill that reads CSV data, \
...  calculates month-over-month growth, aggregates by dimension, \
...  and outputs both a Markdown report and an Excel summary

Use it:

$ zhi run monthly-report january-sales.csv

[file_read] Reading: january-sales.csv (25.6KB)
[file_write] Writing: monthly-report-2026-01.md (3.8KB)
[file_write] Writing: monthly-summary-2026-01.xlsx (5.2KB)

Files saved:
  zhi-output/monthly-report-2026-01.md
  zhi-output/monthly-summary-2026-01.xlsx

Result: A complete analysis report and Excel data table generated with a single command. Fixed format, consistent methodology, ready for management review.

VI. Developer Workflows

6.1 Code Review with Custom Skill

Scenario: You want a quick automated review of source code before committing.

Steps:

Create a code review skill:

name: code-review
description: Review source code and suggest improvements
model: glm-4-flash
system_prompt: |
  You are an experienced code reviewer. Read the provided source code and
  give actionable suggestions. Focus on:
  - Code quality and readability
  - Potential bugs or edge cases
  - Performance issues
  - Security concerns (injection, hardcoded secrets, etc.)
  Output as structured markdown with severity levels.
tools:
  - file_read
  - file_write
max_turns: 10
input:
  description: Source code file to review
  args:
    - name: file
      type: file
      required: true
output:
  description: Code review report
  directory: zhi-output

Run it:

$ zhi run code-review src/auth.py

Result: A structured review report with severity-tagged findings, ready to paste into a pull request comment.

6.2 Generate Test Data

Scenario: You need realistic test data for development or demos.

Steps:

$ zhi -c "Generate 50 rows of realistic test data for a user database: \
  name (Chinese), email, phone (Chinese format), registration date (2024-2026), \
  account status. Save as CSV."

[file_write] Writing: test-users.csv (4.2KB)

File saved: zhi-output/test-users.csv

Result: A CSV file with realistic, locale-appropriate test data. No real personal information is used.

6.3 Summarize Git History

Scenario: You need to write release notes or a changelog based on recent commits.

Steps:

$ git log --oneline --since="2026-02-01" | zhi -c "Organize these commits into a changelog grouped by category (features, fixes, chores)"

Result: A structured changelog with commits categorized and described in user-friendly language.

For a more detailed version:

$ git log --format="%h %s (%an, %ar)" --since="last month" | zhi -c "Write release notes for v2.5 based on these commits"

6.4 Cost-Effective Batch Processing

Scenario: You need to process a large number of files but want to keep costs down.

Cost comparison

Skills use glm-4-flash by default, which costs approximately 10% of GLM-5 chat pricing. For batch operations, this difference adds up significantly.

Strategy: Use skills instead of interactive chat for repetitive tasks:

# Process 20 documents at ~10% of chat cost
for f in docs/*.md; do
  zhi run summarize "$f"
done

# Or use the daily-digest composite skill for a single combined report
zhi run daily-digest ./docs/

Cost breakdown example:

Method Model Per-document cost 20 documents
Interactive chat GLM-5 ~0.05 CNY ~1.00 CNY
Skill execution GLM-4-flash ~0.005 CNY ~0.10 CNY

Skills are the preferred approach for any repeatable, file-based workflow.

Quick Reference

Task Command
Start interactive mode zhi
One-shot question zhi -c "your question"
Pipe input cat file.txt \| zhi -c "summarize this"
Run a skill zhi run <skill> [files...]
First-time setup zhi --setup
Check version zhi --version
Debug mode zhi --debug

Slash Commands in Interactive Mode

Command Description
/help Show help
/model glm-5 Switch to GLM-5 (more capable, higher cost)
/model glm-4-flash Switch to GLM-4-flash (faster, cheaper)
/auto Auto mode (skip tool confirmation prompts)
/approve Approve mode (default, confirm each tool call)
/think Enable thinking mode (GLM-5 only)
/fast Disable thinking mode
/run <skill> [file] Run a skill from interactive mode
/skill list List installed skills
/skill new Create a new skill
/reset Clear conversation history
/undo Undo the last exchange
/usage Show token usage for the current session
/exit Exit zhi

Safety Reminders

  • All file writes go to zhi-output/ -- your original files are never modified.
  • Shell commands always require manual confirmation, even in /auto mode.
  • Destructive commands (rm, mv, etc.) trigger extra warnings.
  • Catastrophic commands (rm -rf /, etc.) are permanently blocked.
  • OCR files are limited to 20MB.
  • Skill names only allow letters, digits, hyphens, and underscores.