Skip to content

cli.md

Latest commit

 

History

History
130 lines (104 loc) · 3.55 KB

cli.md

File metadata and controls

130 lines (104 loc) · 3.55 KB

CLI Reference

AgentForge exposes a top-level agentforge command and an interactive TUI command set.

Top-Level Commands

agentforge

Start the interactive TUI.

agentforge

agentforge run [PROMPT]

Run a single prompt or start interactive mode if no prompt is provided.

agentforge run "summarize this repository"
agentforge run --cwd /path/to/project "inspect the code"

agentforge init

Run the setup wizard. It writes project config and optional environment files with private permissions.

agentforge init

agentforge doctor

Check local configuration and runtime readiness without calling a model.

agentforge doctor
agentforge doctor --json
agentforge doctor --cwd /path/to/project

agentforge report

Print a saved session report without calling a model.

agentforge report
agentforge report --json
agentforge report --session-id <session_id>
agentforge report --data-dir /path/to/data

agentforge completion

Print shell completion setup.

agentforge completion bash
agentforge completion zsh
agentforge completion fish

Interactive Commands

Command Purpose
/help Show TUI commands
/exit, /quit Exit
/new Reset the current session
/clear Clear conversation history
/reload Reload config from disk
/version Show AgentForge version
/retry Retry the last user message
/history [n] Show recent messages
/config Show redacted config
/doctor Show doctor report
/doctor fix Apply safe doctor fixes, such as .gitignore entries and private config permissions
/provider [name] Show or switch provider for this session
/models [--page N] [--limit N] List model suggestions for the current provider
/model list Alias for /models
/model [name] Show or change the model for this session
/fallbacks Show or edit fallback models
/paths Show config, env, session, checkpoint, skill, and cwd paths
/compact Force context compaction
/errors [n] Show recent model/tool errors
/approval <mode> Change approval policy
/tools List registered tools
/skills List available skills
/skill <name> Activate a skill manually
/unskill <name> Deactivate a skill
/mcp Show MCP server status
/name [name] Show or set session name
/save Save session snapshot
/sessions [--page N] [--limit N] List saved sessions
/resume <session_id> Resume a saved session
/checkpoint Save a checkpoint
/checkpoints [--page N] [--limit N] List checkpoints
/restore <checkpoint_id> Restore a checkpoint
/plan Switch to read-only plan mode
/build Switch to normal build mode
/todos Show todos
/todos --clear Clear todos
/stats Show session stats
/report Show current session report
/report --json Print current session report as JSON
/export Export session as markdown
/export html Export session as HTML

Suggested First Session

/doctor
/tools
/skills
Read README.MD and explain the architecture.
/plan
Suggest a small improvement without editing files.
/build
Create a small notes file summarizing the project.
/checkpoint
/export html
/stats

Approval Tips

  • Use on-request for normal development.
  • Use never for read-only exploration.
  • Use auto-edit when you trust local edits but still want protection around riskier operations.
  • Avoid yolo outside disposable sandboxes.