Skip to content

getting-started.md

Latest commit

 

History

History
100 lines (72 loc) · 3.4 KB

getting-started.md

File metadata and controls

100 lines (72 loc) · 3.4 KB
Raw

Getting Started

Build and Run

packetcode requires Go 1.24.2 or newer.

make build
./bin/packetcode

The first run creates ~/.packetcode/config.toml. Choose a provider, enter an API key when required, and select a model.

For a Codex ChatGPT subscription, sign in through the official Codex CLI first:

codex login
packetcode --provider codex

For local Ollama, start the standard daemon; packetcode uses http://localhost:11434 with no key or extra configuration:

ollama serve
packetcode --provider ollama

Useful startup commands:

packetcode
packetcode --provider codex --model gpt-5.6-sol
packetcode --resume <session-id>
packetcode --permission-mode auto
packetcode --trust
packetcode doctor
packetcode doctor --json

--provider requires an already-configured provider, except keyless Codex/Ollama. Use Ctrl+P, /provider, or /provider add <slug> to add a hosted provider key.

Everyday Keys

Key Action
Enter Send.
Alt+Enter / \ then Enter Portable newline; Ctrl+J also works while completion is closed, and Shift+Enter works when distinctly reported.
Up / Down Prompt history at the input boundary.
Shift+Tab Cycle Manual → Accept Edits → Auto → Plan, even during a turn.
Left on an empty prompt Open Agent View.
Ctrl+P / /model Provider/model picker; Ctrl+M is terminal-dependent.
Ctrl+C Cancel active work, clear a draft, or quit from an empty prompt.
Ctrl+L Clear visible output, keep the saved session.

Finalized output goes to terminal scrollback. The bottom live region contains only current thinking, streaming text, tool output, input, context status, and permission mode.

Prompts and Context

  • Type / for command completion.
  • Type @ at a token boundary for fuzzy project-file completion.
  • Submit during an active turn to queue the prompt.
  • Use //text to send a prompt beginning with a literal /.
  • Use /compact to summarize older history manually; automatic compaction also runs at the configured threshold.

The context gauge is current occupancy, not cumulative billed tokens.

Approvals and Modes

The numbered approval menu supports arrows, Enter, 1/2/3, and the legacy Y/A/N shortcuts. Option 2 remembers a session rule; shell approvals remember the exact command.

Shift+Tab remains active while the model is thinking or streaming. If an approval is already visible, switching to a mode that allows or denies the action resolves it immediately. Already-running processes are not interrupted.

Bypass Permissions is intentionally separate from the Shift+Tab cycle:

packetcode --trust

or /trust on for the current session. Explicit deny rules remain effective.

Agents, Loops, and Workflows

/spawn audit the current diff
/spawn --write fix the focused tests
/agents
/workflows run review target="the staged diff"
/loop Continue improving the implementation until complete
/loop 15m /workflows run review

Read-only agents share the project root. Write-capable agents receive a dedicated git worktree. Results remain outside foreground model context until explicitly collected or injected.

Where Next