Skills Reference¶

zhi ships with 15 built-in skills organized into two categories: single-purpose skills that perform one task, and composite skills that orchestrate multiple skills together.

Single-Purpose Skills¶

summarize¶

Summarize a text file or document into a concise markdown overview.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `file_write`
Max turns	5

Usage:

zhi run summarize report.txt

Output: A markdown summary saved to zhi-output/ containing key points, main arguments, and important details.

translate¶

Translate text between languages, defaulting to Chinese (Simplified).

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `file_write`
Max turns	5

Arguments:

Name	Type	Required	Default	Description
`file`	file	No	-	File to translate
`text`	string	No	-	Text to translate
`to`	string	No	`chinese`	Target language

Usage:

zhi run translate readme-en.md

extract-text¶

Extract text from PDF or image files using OCR and save as plain text.

Property	Value
Model	`glm-4-flash`
Tools	`ocr`, `file_write`, `file_list`
Max turns	5

Usage:

zhi run extract-text invoice.pdf

Output: A cleaned .txt file preserving document structure (headings, paragraphs, lists). Page headers/footers are removed, and tables are rendered as aligned text.

extract-table¶

Extract tables from documents and save as CSV or Excel.

Property	Value
Model	`glm-4-flash`
Tools	`ocr`, `file_read`, `file_write`, `file_list`
Max turns	8

Usage:

zhi run extract-table report.pdf

Output: One or more CSV files with proper column headers. Multiple tables are saved as separate files (table1.csv, table2.csv) or combined if they share the same schema.

Excel output

Request .xlsx output in your prompt and the skill will produce Excel files instead of CSV.

analyze¶

Deep analysis of a document with structured insights.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `file_list`
Max turns	8

Usage:

zhi run analyze proposal.md

Output: A markdown report with four sections:

Document Overview -- type, length, structure, primary language
Key Findings -- main topics, arguments, data points
Structure Analysis -- organization, logical flow, gaps
Actionable Takeaways -- implied decisions, open questions, follow-up steps

proofread¶

Proofread a document and output a correction report.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`
Max turns	5

Usage:

zhi run proofread draft.md

Output: A markdown report containing:

Overall quality score (out of 10)
Per-issue corrections with location, original text, corrected text, and error type (spelling / grammar / punctuation / style / clarity)
Summary of recurring patterns

reformat¶

Convert documents between formats.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `file_list`
Max turns	8

Arguments:

Name	Type	Required	Default	Description
`file`	file	Yes	-	File to convert
`format`	string	No	`markdown`	Target format

Supported conversions:

From	To
Plain text	Markdown
Markdown	Plain text
CSV	Markdown table
CSV	Excel (`.xlsx`)
Any text	CSV
Any text	JSON
PDF/Image (via OCR)	Markdown / Text / CSV

Usage:

zhi run reformat data.csv

meeting-notes¶

Structure raw meeting notes into organized minutes.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`
Max turns	5

Usage:

zhi run meeting-notes notes.txt

Output: Structured markdown meeting minutes containing:

Attendees -- list of participants
Agenda Items -- key discussion points and decisions per topic
Action Items -- table with task, owner, and deadline
Key Decisions -- numbered list
Next Steps -- follow-up items and next meeting date

compare¶

Compare two documents and highlight differences.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `file_list`
Max turns	8

Arguments:

Name	Type	Required	Description
`file`	file	Yes	First file
`file2`	file	Yes	Second file

Usage:

zhi run compare v1.md v2.md

Output: A markdown diff report with comparison summary, added/removed/modified sections, and unchanged sections. Focuses on meaningful differences and ignores whitespace-only changes.

Composite Skills¶

Composite skills orchestrate multiple single-purpose skills using skill_<name> tool wrappers. They have higher turn limits to accommodate multi-step workflows.

contract-review¶

Multi-step contract analysis combining analysis, comparison, and proofreading.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `file_list`, `skill_analyze`, `skill_compare`, `skill_proofread`
Max turns	15

Arguments:

Name	Type	Required	Description
`file`	file	Yes	Contract to review
`file2`	file	No	Previous version for comparison

Usage:

zhi run contract-review contract-v2.pdf contract-v1.pdf

Output: A comprehensive review report containing:

Executive summary with overall assessment
Structural analysis (key clauses, obligations, risks)
Version comparison (if previous version provided)
Language and consistency issues
Risk assessment (liability, indemnity, termination)
Numbered recommendations

daily-digest¶

Batch-summarize all documents in a folder into a single digest.

Property	Value
Model	`glm-4-flash`
Tools	`file_list`, `file_read`, `ocr`, `file_write`, `skill_summarize`
Max turns	15

Supported file types: .txt, .md, .pdf, .csv, .docx, .xlsx, .png, .jpg

Usage:

zhi run daily-digest ./reports/

Output: A single markdown digest with:

Overview (document count, key themes)
Individual summaries per document
Cross-document insights (common themes, contradictions, follow-up actions)

invoice-to-excel¶

OCR invoices and consolidate line items into an Excel spreadsheet.

Property	Value
Model	`glm-4-flash`
Tools	`file_list`, `file_read`, `ocr`, `file_write`, `skill_extract-table`, `skill_reformat`
Max turns	15

Usage:

zhi run invoice-to-excel invoices/

Output: An Excel file with:

Sheet 1 "Line Items": Invoice #, Date, Vendor, Item, Quantity, Unit Price, Amount, Tax, Total
Sheet 2 "Invoice Summary" (multiple invoices): Invoice #, Date, Vendor, Subtotal, Tax, Total, Status

Data normalization

Currencies are normalized (symbols removed, consistent decimals). Dates are formatted as YYYY-MM-DD. Missing fields show N/A.

meeting-followup¶

Generate structured minutes, executive summary, and optional translation from raw meeting notes.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `skill_meeting-notes`, `skill_summarize`, `skill_translate`
Max turns	15

Arguments:

Name	Type	Required	Description
`file`	file	Yes	Raw meeting notes
`translate_to`	string	No	Target language for summary translation

Usage:

zhi run meeting-followup standup-notes.txt

Output: Up to three files:

[meeting]-minutes.md -- full structured meeting minutes
[meeting]-summary.md -- executive summary (1 page max)
[meeting]-summary-en.md -- translated summary (if applicable)

report-polish¶

Proofread, analyze structure, and produce a polished final version of a document.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `skill_proofread`, `skill_analyze`, `skill_reformat`
Max turns	12

Arguments:

Name	Type	Required	Default	Description
`file`	file	Yes	-	Draft document
`format`	string	No	`markdown`	Output format (markdown, docx, txt)

Usage:

zhi run report-polish draft-report.md

Output: Two files:

[filename]-final.[ext] -- polished document
[filename]-changes.md -- change log with before/after quality scores, language fixes, structural improvements, and remaining suggestions

translate-proofread¶

Translate a document and proofread the translation for quality.

Property	Value
Model	`glm-4-flash`
Tools	`file_read`, `ocr`, `file_write`, `skill_translate`, `skill_proofread`
Max turns	12

Arguments:

Name	Type	Required	Default	Description
`file`	file	Yes	-	Document to translate
`to`	string	No	`chinese`	Target language

Usage:

zhi run translate-proofread whitepaper.md

Output: Two files:

[filename]-translated.[ext] -- polished final translation
[filename]-translation-report.md -- quality report with detected source language, quality score (1-10), issues fixed, and passages needing human review

Creating Custom Skills¶

You can create custom skills by writing YAML configuration files. Place them in your user skills directory or use the skill_create tool.

name: my-skill
description: What this skill does
model: glm-4-flash
system_prompt: |
  Your instructions to the model...
tools:
  - file_read
  - file_write
max_turns: 10

Skill naming rules

Skill names must match ^[a-zA-Z0-9][a-zA-Z0-9_-]*$ and be at most 64 characters long. No spaces, slashes, or special characters.