Module check_runner 

Task checks the node runs itself. When a session’s task worktree (task-<id>) has a new commit and a clean tree, the node asks core for the task’s check command (task.check_spec), runs it in that worktree and reports the exit code (task.check_run, signed by the node’s key). Core takes the receipt as proof, not the agent’s word; the same path serves PTY and ACP sessions.

The command runs code from the agent’s tree (tests, build scripts), so it runs only through exec-as with the node’s run-as user, exactly like the sessions; a node without --run-as (and Windows) does not run checks. The check sees only ENV_KEEP of the node’s variables, gets no stdin, its output is dropped, and its process group is killed at the timeout.

Constants

ENV_KEEP

The only node variables a check sees: tool paths and locale, never the node’s own settings (core address, token, m4a keys, provider credentials).

TIMEOUT

Longest a check may run; past it the check fails (exit code -1).

Functions

argv

The program and arguments that run command through the shell (and the run-as user).

due

Task worktrees with a new commit and a clean tree whose head was not checked yet.

kept_env

run

Run command in root: Some((exit_code, duration_ms)), -1 for a timeout or a check that could not start. None: not run, the node has no --run-as user (a check is a command from core and never runs as the node’s own user).

Type Aliases

Due

A worktree whose head wants a check: (workspace, root, head).