Skip to content

README.md

Latest commit

 

History

History
43 lines (36 loc) · 2.8 KB

README.md

File metadata and controls

43 lines (36 loc) · 2.8 KB
Raw

src/cli/

Everything that turns the agent core into an interactive terminal chat. main.py at the repo root is a thin entry point — it wires up skills/MCP discovery at startup and then hands off to this package for the actual input loop.

  • display.py - ANSI colors, the startup banner, Spinner (the animated AI ❯ thinking... status line shown while the agent works), the arrow-key selector (select_one, used by the startup online/offline prompt and the /models picker), and the block-layout helpers every command prints through (print_heading, print_rows, print_fields, column_width) so padding and color are decided in one place.
  • session.py - Session — owns the active model/agent, chat history, and history-trimming rules (trim_history, compact_turn, keep_only_latest_image) that keep the rolling conversation under the model's context budget.
  • commands.py - one cmd_* function per slash command (/help, /new, /clear, /purge, /memory, /status, /models — an arrow-key picker, with default/info variants — /network and /network info, /image, /tools, /skills, /mcp) plus handle_command(), the dispatch table main.py's input loop calls. apply_network_mode() lives here too: it is the one place the live mode, the persisted setting, and memory's embedder policy are changed together, and startup calls it for exactly that reason.
  • turn.py - run_turn(): builds the message list for one user turn, drives src.engine.turn_engine.run_turn(), and renders each TurnEvent as it arrives (spinner label, trace lines, live answer tokens) before folding tool-call intermediates back out of history once the turn completes.

Adding a new slash command

  1. Write a cmd_<name>(session, ...) function in commands.py that prints whatever it needs to.
  2. Add a branch to handle_command() matching /<name>.
  3. Add a row to _HELP_SECTIONS under the group whose subject it acts on (Chat, Models, Network, Memory, Inspect, Session, or a new one) so /help picks it up automatically.
  4. Print through display.py's helpers rather than hand-rolled print() formatting, so the new command lines up with the existing ones (/help and /network info pad their name and label columns to one width across every block, which only works if all output goes through them).

No other file needs to change — main.py only ever calls handle_command(), never a specific command function.

Why this is split out

Nothing in this package is imported by src/engine/ or the tool/model/skill registries — only main.py reaches into it. That keeps the agent core and its registries usable headless (e.g. a future web or API frontend) without dragging in print()/input() terminal code.