v1.7.9

v1.7.9 makes a Cloud Agent run something you can follow from the first minute instead of a wait for a final pull request.

#The draft pull request opens first

A run used to open its pull request only after the agent finished, so there was nothing to look at while it worked. The draft pull request now opens before the agent starts, on a branch with a single "Initial plan" commit, and the person who asked is assigned to it. The commits the agent pushes land on it as they go, and its description lists them as they arrive. When the run ends, the description is replaced with the agent's summary. A run that finishes without committing anything closes its draft pull request, and the failure comment says so.

#Commits are pushed as the agent works

The platform now pushes the agent's commits during the run, not only at the end, so a run that is cut off still ships what it finished. What the agent left behind is salvaged and pushed rather than lost. This is what fills the draft pull request while the run is still going.

#The run page streams the transcript

The run page now streams the transcript live while the agent works, so you can watch a run from sigit.si instead of waiting for it to land on the pull request.

#Clearer reasons when a run stops short

When a run stops before it is done, the message now says what actually happened, so a run that hit its wall clock, its tool-round limit, or a cancel reads as that rather than as a generic failure.

#A Ruby toolchain in the sandbox for this repository

The agent's sandbox provides a generic substrate (bwrap, git, curl and python3) and nothing language-specific. The sigit.si repository now installs its own Ruby into the sandbox during setup, from its .sigit/code-cloud-agent/config.toml, so a run here can install gems and run the suite. Other repositories declare whatever toolchain they need the same way.