AVL Code

6. The five work modes

Work modes are the core of AVL Code's assistant system. Each workspace binds a different persona to each of the five modes.

6.1 Mode overview

Mode Icon Default behaviour Typical use
auto One-shot answers with full tool access Quick Q&A, simple tasks
plan Read-only: no writes / no command execution Design, code reading, risk review
prepare Organise requirements, set up dependencies Project bootstrap, checklist generation
execute Write code, run commands, commit changes The actual work
assess No code changes: can run tests and checks, draws conclusions Acceptance, retrospective; no further edits

plan is a true read-only mode: write-file, edit and execute tools are all disabled — it can't even look at background command output. The point is to keep the "think it through" stage pure. The restriction lives at the tool layer; a persona configuration can only make it stricter, never loosen it.

assess doesn't write project code, but it can run commands: so that acceptance genuinely gets done, it can run tests, lint and SAST, generate an SBOM, and do security analysis and threat-intelligence lookups. What it can't do is write or edit files. If what you want is "touch absolutely nothing", use plan.

6.2 Switching modes

Five round buttons below the input box switch modes with one click. The switch automatically:

  1. Loads the mode-bound persona.
  2. Applies the mode's tool-enable set.
  3. Inserts a "mode switch" marker card in the message stream for traceability.

6.3 Mode relationships

Modes form a recommended workflow:

plan → prepare → execute → assess

auto is the catch-all for small tasks that don't need strict staging. AVL Code encourages plan first, then execute: the read-only constraint in plan stops the AI rushing into code while misunderstanding the premise.

6.4 Exiting plan

Once the plan looks good, click the "Exit Plan, Enter Execute" banner at the top of the conversation. AVL Code switches to execute and injects the confirmed plan into the next turn.

6.5 Customising mode behaviour

Per mode you can configure:

  • Default persona
  • Tool switches (allow / ask / deny per tool)
  • Model and parameters
  • Auto-accept / auto-deny thresholds (advanced)

Adjust per-workspace under Settings → Agents.