---

name: cli-and-conversations description: "Operate gptme from the terminal: CLI prompts, conversation logs, slash commands, automation, and persistent agent commands." disable-model-invocation: true metadata: disco-role: operating license: MIT

cli-and-conversations

Use this sub-skill when the task is about running gptme from a terminal, shaping prompts, managing chat histories, using slash commands, queuing follow-up prompts, building non-interactive automation, or understanding the gptme-agent management CLI.

Route away when the task is mainly about:

• provider credentials, default model choice, OAuth, or config precedence: use the configuration/providers sub-skill.
• built-in tool internals, custom tools, plugins, hooks, skills, lessons, MCP, browser, or computer-use setup: use the tools/extensibility sub-skill.
• gptme-server, REST/SSE, Web UI, ACP, TUI, deployment, or hosted access: use the server/protocols sub-skill.
• evaluation suites, SWE-bench/T-bench, Docker benchmark runs, or leaderboard processing: use the evals/benchmarks sub-skill.
• maintaining a gptme checkout, running the contributor test suite, releases, or Web UI development: use the repo-development sub-skill.

Read first

• references/cli-reference.md for install/start patterns, main CLI options, prompt/context syntax, slash-command map, tool-selection syntax, and gptme-agent command families.
• references/conversations-and-automation.md for named conversations, resume/search/fork, multiprompts, queued prompts, non-interactive automation, log layout, and persistent agent workflows.
• references/troubleshooting.md for missing explicit local paths, resume/fork confusion, non-interactive pitfalls, tool allowlist errors, stdin behavior, context minimization, and agent service issues.

Safe helpers

• scripts/build_gptme_command.py builds a shell-quoted gptme command from explicit prompt/model/tool/context/name flags without executing it.
• scripts/inspect_conversation_log.py summarizes and redacts a conversation.jsonl file, one conversation directory, or a directory of conversation directories without importing gptme.

Fast operating checklist

1. Confirm the user wants terminal CLI/conversation behavior, not provider or tool internals.
2. Prefer named conversations for resumable work: gptme --name <conversation> "task".
3. Use the standalone - argument only to separate chained turns; a lone - by itself is a literal prompt, not a stdin marker.
4. For tight startup context, combine --system short, a narrow --tools value, and either --context files or --no-workspace.
5. For automation, use --non-interactive; add --output-format json only with non-interactive mode.
6. If diagnosing logs, inspect an explicit log path with the bundled inspector before assuming which conversation --resume will choose.