aricode v0.5.1: section mode gets practical

v0.5.0 was the ambitious release. It added dreaming, the public Node SDK, behavioral compilation, and the rebrand to aricode. It also left a few rough edges in the day-to-day operator experience. v0.5.1 is the pass where those edges get filed down.

Section mode now behaves like a planner, not a suggestion

The biggest practical change is in /sectmode. The master planner now recovers from malformed dispatch JSON, retries when a dispatch block is present but unusable, and resolves section ids from normalized aliases instead of requiring one exact spelling every time.

That matters because the previous failure mode was maddeningly simple: the planner could clearly describe the sections it wanted to activate and still fall through to “no sections need changes” because the dispatch block was slightly malformed or too human-looking.

In 0.5.1, that path is much more forgiving:

• Indented ```dispatch fences are accepted.
• Dispatch blocks with newline-broken JSON strings get sanitized and retried.
• Section ids can resolve from exact ids, normalized ids, and slugified labels.
• If the master has already attempted a dispatch, aricode no longer quietly pretends that no work is needed.

Basic tasks stop burning so many turns

The second fix is more subtle but just as important: turn budgets. The first sectmode pass was too generous. The master planner, section agents, and verifier could spend far too many turns on a fairly ordinary task, which made the whole system feel sluggish even when it was technically working.

v0.5.1 cuts those ceilings aggressively enough that “implement this feature” feels like a coding task again instead of a long orchestration ceremony. The result is lower latency, less token waste, and fewer runs that over-explore before they commit to work.

The Ink REPL now acts more like a terminal and less like a demo

The UI pass in this release is about behavior, not just styling. The REPL now favors a transcript-first flow, keeps normal scrollback in the main terminal buffer, and avoids the kind of mouse capture and alternate-screen side effects that made copy and selection feel hostile.

That translates to a few practical wins:

• Selections are copyable again.
• Scrollback works like a normal CLI transcript.
• The input path is less fragile.
• Status and transcript rendering are calmer, with fewer unnecessary repaints.

Section Mode is now the default way to scale work

This release reframes /sectmode from “promising but fragile” to “feature-complete enough for real implementation loops.” Dispatch retries, alias-aware targeting, and lower orchestration ceilings combine into a workflow that behaves like engineering tooling, not a lab demo.

/sectmode init
/sectmode on

# ask for a real feature split across sections
Implement the new auth settings panel and wire tests.

That means less babysitting of planner output, fewer dead-end dispatches, and faster execution on tasks that should have been routine in the first place.

If you used early section mode and bounced off it, v0.5.1 is the release to revisit.

Why this release matters

There is a category of release that adds more capabilities. And there is a category that makes the capabilities you already shipped feel real. v0.5.1 is the second kind.

It does not try to outshine the dreaming or SDK launch. It makes the operator loop tighter: planner dispatch is resilient, section routing is more accurate, and the terminal gets out of your way while long tasks run.

Get the update

If you already use aricode, update and start a fresh REPL session. Then enable section mode and run a real implementation task across two or more sections to feel the difference immediately.

For the full reference, head to the documentation. For the release-by-release history, see the changelog. And if you want the bigger launch story behind the current direction, the v0.5.0 post is still the place to start.