> ## Documentation Index
> Fetch the complete documentation index at: https://gump.build/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Provider Quirks

> Each agent CLI has its own behavior, flags, and limitations.

# Provider Quirks

Each agent CLI has its own behavior, flags, and limitations. Gump's adapters handle the differences, but understanding them helps when debugging or choosing agents.

## Claude Code

The most full-featured adapter. Supports all session modes, reports cost natively, granular token breakdowns per message. Uses `--dangerously-skip-permissions` for headless mode. `--max-turns`, `--resume <uuid>`, `--cwd <path>`.

## Codex

Sandbox mode with network disabled by default. `--full-auto`. Token reporting per turn (less granular). Shell classification is reconstructed (lower confidence). `resume <thread-id>`.

## Gemini CLI

Auto-approve mode via `--yolo`. Session resume via `--resume` (implicit, per-directory). No `--cwd` flag — controlled via `exec.Cmd.Dir`. Token reporting aggregated in final `result` event.

## Qwen CLI

Similar to Claude. `--yolo` for auto-approve. `--max-session-turns <N>`. No `--cwd` flag. `--resume <uuid>`.

## OpenCode

The most different adapter. Uses file-backed stdout (blocks on `StdoutPipe()`). Native `--dir` flag. `--session <ses_id>`. Token reporting per `step_finish`. The only provider where exit code 1 is meaningful.

## Cursor Agent

Separate binary from the IDE (`cursor-agent`, auto-installed on first `cursor agent`). Uses `-p`/`--print` for headless mode. `--yolo` for auto-approve. `--trust` required in headless (blocks otherwise). `--workspace <path>` for working directory. `--resume [chatId]`. No `--max-turns` flag — guard must count turns from stream. No native cost reporting — estimated from tokens. Context via `.cursor/rules/*.mdc` files (not CLAUDE.md). camelCase token fields in output (`inputTokens`, not `input_tokens`).

## Cross-provider limitations

Session resume across agents is blocked at the engine level. Only Claude Code reports cost natively — all other providers (including Cursor) use estimated cost from token counts. Each provider has different tool names mapped internally.

## Provider volatility

Agent CLIs are moving fast. If a provider breaks its structured output, Gump falls back to compat mode (raw stdout + git diff + fresh sessions). The workflow still runs — metrics become less granular.