This repository contains runnable example workflows for Cori. The examples are meant to be useful starting points: small enough to read, but realistic enough to adapt.
From this repository, validate a workflow before running it:
cori check ./hn_digestRun a workflow by passing parameters as name=value pairs:
cori run ./hn_digest story_count=5Authenticated workflows may require one-time login first:
cori login gh
cori login gws
cori login openaiAuth-free capability binaries only need a one-time install:
cori capability install anydoc # document conversion (needs Node.js)
cori capability install lightpanda # headless browser (Linux/macOS)| Workflow | Level | Capabilities | Cost | What it shows |
|---|---|---|---|---|
hello_world |
Beginner | curl, code |
Free | End-to-end demo: fetch, transform, format |
code_only |
Beginner | code |
Free | Pure TypeScript step chaining |
hn_digest |
Beginner | curl, code |
Free | Public HTTP APIs and formatting |
github_release_notes |
Beginner/intermediate | gh, llm |
LLM tokens | GitHub API reads plus a focused LLM drafting step |
changelog_draft |
Intermediate | git, llm, code |
LLM tokens | Local git history, diff stats, and stable changelog markdown rendering |
gcal_daily_brief |
Intermediate | gws, llm |
LLM tokens | Calendar read, Gmail send, and typed email assembly |
doc_to_markdown |
Beginner | anydoc, code |
Free | Any office document (Word, PowerPoint, Excel, PDF, …) converted to clean Markdown |
web_page_digest |
Beginner | lightpanda, code |
Free | Headless-browser crawl (JavaScript executed) with structured data extraction |
stale_issue_triage |
Intermediate/advanced | gh, code |
Free | GitHub maintenance with deterministic filtering and dry-run behavior |
security_ownership_map |
Advanced | python3, git |
Free | Security-oriented ownership analysis, orphaned sensitive code, and bus-factor hotspots |
drive_doc_summarizer |
Advanced | gws, llm |
LLM tokens | Drive export, Docs creation, Docs update, and Gmail delivery |
Each workflow is a folder with:
<workflow>/
manifest.md
types.ts # optional shared zod schemas
steps/
01_first_step.ts
02_second_step.ts
README.md # workflow-specific setup and run notes
The numbered files in steps/ are the execution order. Every step exports exactly one step.code, step.cli, or step.llm definition from @cori-do/sdk.
Start with hello_world if you want the fastest no-auth smoke test that proves everything works.
Use github_release_notes when you want to see a practical LLM step that turns structured API data into publishable text.
Use changelog_draft when you want to turn a local git release range into Highlights, Changes, and Fixes markdown for a changelog draft.
Use gcal_daily_brief or drive_doc_summarizer when you want to adapt Google Workspace automation.
Use stale_issue_triage when you want a deterministic maintenance workflow that is safe to run in dry_run=true mode.
Use security_ownership_map when you want to analyze a repository's security-sensitive ownership, uncover hidden owners, and surface low-bus-factor hotspots from git history.
security_ownership_map is a direct workflow transformation of OpenAI's official security-ownership-map skill.
changelog_draft is adapted from @openclaw/openclaw-changelog-update, an agent skill for generating OpenClaw release changelogs from git history.
Using the reported token counts as input / output, the original agentic flow used about 43k input / 176k output tokens, while this Cori workflow used about 3.6k input / 78 output. That is a reduction of about 39.4k input tokens (91.6%) and 175,922 output tokens (99.96%), or roughly 215,322 fewer tokens overall (98.3%).
Using standard API pricing checked on 2026-06-06 from Anthropic, OpenAI, and Google Cloud, the same token profile works out to approximately:
| Model | Input / output price per 1M | Agentic flow | Cori workflow | Savings |
|---|---|---|---|---|
Claude Opus 4.8 |
$5 / $25 |
$4.6150 |
$0.0200 |
$4.5951 (99.57%) |
Claude Sonnet 4.6 |
$3 / $15 |
$2.7690 |
$0.0120 |
$2.7570 (99.57%) |
GPT-5.4 |
$2.50 / $15 |
$2.7475 |
$0.0102 |
$2.7373 (99.63%) |
GPT-5.5 |
$5 / $30 |
$5.4950 |
$0.0203 |
$5.4747 (99.63%) |
Gemini 3.5 Flash |
$1.50 / $9 |
$1.6485 |
$0.0061 |
$1.6424 (99.63%) |
These examples use standard pricing only, not cached-input, batch, flex, or priority rates. The Gemini figure uses the official Gemini 3.5 Flash global standard tier for requests with <= 200K input tokens.