kodelet

Kodelet is a lightweight agentic software-engineering CLI. This skill gives a quick orientation; deeper topics live in focused references.

Reference map

For details, choose the reference that matches the task:

• references/quick-start.md — installation, core kodelet run usage, Web UI, ACP/IDE mode, image input, git commands, completions, common workflows.
• references/conversations.md — conversation list/show/delete/fork commands, renaming, steering, and the ACP Streamlit example.
• references/configuration.md — config files, profiles, model/provider setup, security restrictions, MCP, bash timeout.
• references/recipes.md — AGENTS.md, fragments, and recipes.
• references/skills.md — agentic skills, skill layout, configuration, and skill plugins.
• references/extensions.md — extension runtime, discovery, configuration, plugins, persistent TUI widgets/surfaces, and operational behavior.
• references/sdk.md — TypeScript SDK agent sessions, extension authoring API, tools, commands/dynamic recipes, native TUI shortcuts, lifecycle events, UI helpers, persistent surfaces, and examples.

Examples now live with the skill under examples/:

• examples/streamlit-acp/ — Streamlit chatbot using Agent Client Protocol.
• examples/extensions/workspace/ — TypeScript extension with an ask_user_question tool and bash approval policy.
• examples/sdk/ — TypeScript SDK examples for basic, streaming, and inline-extension agent sessions.

When answering detailed or version-sensitive questions, prefer the current repository docs/source over memory. If the user asks to modify Kodelet itself, inspect the relevant code first and follow repository conventions from AGENTS.md.

Essential commands

# One-shot agent run
kodelet run "your query"

# Continue the most recent conversation
kodelet run -f "continue the task" # same as --follow

# Resume a specific conversation
kodelet run --resume CONVERSATION_ID "more questions"

# Temporary or minimal runs
kodelet run --no-save "temporary query"
kodelet run --result-only "what is 2+2"
kodelet run --no-tools "what is the capital of France?"

# Interactive/IDE mode over Agent Client Protocol
kodelet acp

# Browser UI
kodelet serve

Feature summary

• Context files: Kodelet automatically loads AGENTS.md; bootstrap one with kodelet run -r init.
• Recipes/fragments: User-invoked prompt templates in ./recipes/ or ~/.kodelet/recipes/; list with kodelet recipe list, inspect with kodelet recipe show <name>, run with kodelet run -r <name>.
• Skills: Model-invoked domain guidance in .kodelet/skills/<name>/SKILL.md, plugins, or global skill dirs; disable with --no-skills.
• Extensions: Long-running subprocesses can register model tools, prompt commands/dynamic recipes, native TUI shortcuts, and lifecycle event handlers; inspect with kodelet extension list and disable with --no-extensions.
• Plugins: Install bundled skills, recipes, and extensions with kodelet plugin add org/repo; inspect with kodelet plugin list and kodelet plugin show org/repo.
• Conversations: Use kodelet conversation list/show/delete/fork for persisted runs.
• Git helpers: kodelet commit generates commit messages; kodelet pr creates PRs.

Quick decision guide

• User asks “how do I run/use/install Kodelet?” → use references/quick-start.md.
• User asks about config, providers, models, command/tool restrictions, MCP → use references/configuration.md.
• User asks about recipes, fragments, or AGENTS.md → use references/recipes.md.
• User asks about agentic skills, skill layout, or skill plugins → use references/skills.md.
• User asks about extension discovery/runtime/config, plugins → use references/extensions.md.
• User asks about the TypeScript SDK, SDK agent sessions, SDK extension tools/commands/events, or SDK UI prompts → use references/sdk.md.
• User asks about conversation history, renaming, steering, or the Streamlit ACP example → use references/conversations.md.