Prompting and plan mode

You drive a session by giving it instructions. There are two modes, depending on how much you want the agent to do.

Execute mode (default)

The agent reads your code, makes the changes, and produces a result you can review. Use it for normal work:

chronos session new "refactor the auth module to use the new token type"
chronos session send <id> "add unit tests for the edge cases"

Changes are made inside the isolated session, never on your main branch — so you can always review before anything ships.

Plan mode asks the agent to research and propose a plan without changing anything. Use it to preview an approach before committing to it.

There are three ways to use it. From the command line:

chronos session new --mode plan "how would you migrate this to async?"

In the dashboard, toggle Plan in the composer. Or start a single message with /plan:

/plan walk me through the safest way to split this module

After reviewing the plan, send a normal instruction to carry it out.

Pick the agent when you create a session — in the dashboard’s composer selector, or:

chronos session new --agent <name> "…"

Your organization decides which agents are available.

Be specific about the outcome and any constraints (style, tests, files to avoid).
For anything large or risky, plan first, then execute.
Work in small steps rather than one giant instruction — you get cleaner, reviewable checkpoints.