banana-cli（banana CLI 搜索）

banana-cli

CLI for creating, managing, and exporting AI-generated presentations.

Environment Check

Before running any command, verify the backend is reachable:

curl -sf http://localhost:5000/health

If this fails, the backend is not running. Read references/setup.md and follow the steps to clone the repo, configure .env, and start the backend. Do not proceed until the health check passes.

Invocation

banana-cli <command> [options]

If banana-cli is not on PATH, use uv run banana-cli from the project root, or install globally with uv tool install.

End-to-End Workflow

# 1. Create project and set as working project
result=$(banana-cli --json projects create --creation-type idea --idea-prompt "Your topic")
project_id=$(echo "$result" | jq -r '.data.project_id')
banana-cli projects use "$project_id"

# 2. Generate everything (outline → descriptions → images)
banana-cli workflows full --language zh --pages 8

# 3. Export to local file
banana-cli exports pptx --output ./slides.pptx

Once a working project is set, --project-id is optional on all subsequent commands.

Key Patterns

Short ID prefix matching

All --project-id and --page-id accept short prefixes (like git short hashes):

banana-cli projects get a1b2          # matches a1b2c3d4-...
banana-cli pages edit-image --page-id b9c8 --instruction "change title to red"

Working project context

Avoid repeating --project-id by setting a working project:

banana-cli projects use a1b2     # set (accepts prefix)
banana-cli workflows outline      # uses working project
banana-cli projects use           # show current
banana-cli projects unuse         # clear

Page count control

--pages is a hint to the AI — actual page count may differ. The CLI warns on stderr when they don't match.

banana-cli workflows outline --pages 5
banana-cli workflows full --pages 10 --language en

Export with auto-download

# Download directly to local path
banana-cli exports pptx --output ./slides.pptx
banana-cli exports pdf --output ./report.pdf

# Without --output, returns a server-side download URL
banana-cli exports pptx

Batch generation

cat > jobs.jsonl << 'EOF'
{"job_id":"t1","job_type":"full_generation","creation_type":"idea","idea_prompt":"AI Intro","language":"zh","export":{"formats":["pptx"]}}
{"job_id":"t2","job_type":"full_generation","creation_type":"idea","idea_prompt":"ML Basics","language":"zh","export":{"formats":["pptx","pdf"]}}
EOF

banana-cli run jobs --file jobs.jsonl --report report.json --state-file state.json

Renovate existing PPT

banana-cli renovation create --file /absolute/path/to/slides.pptx --language zh

JSON output for scripting

banana-cli --json projects list | jq '.data.projects[].project_id'

Important Notes

• File path arguments (--file, --image) require absolute paths
• Async tasks (descriptions, images, editable export) wait by default and show progress on stderr. Pass --no-wait to get a task_id immediately
• --wait / tasks wait can be interrupted (Ctrl+C) and resumed anytime — backend tasks are unaffected
• Progress lines go to stderr (format: [PROGRESS] STAGE STATUS completed/total), keeping stdout JSON clean
• --help output is plain text when piped (non-TTY) — safe for agent consumption
• Config priority: CLI args > env vars (BANANA_CLI_*) > TOML config (~/.config/banana-slides/cli.toml) > defaults

Discovering Commands

Run banana-cli --help for the top-level command list, and banana-cli <command> --help for subcommand options. Help output is plain text when piped (non-TTY).