Files
ai-agent-harness/docs/local-scheduling.md

2.0 KiB

Local Scheduled Workspace Verification

The Windows Task Scheduler runs the same read-only workspace verification as the manual command. It does not compose documents, update repositories, perform Git operations, or access the network.

Manual verification with a retained log

Run the scheduled-verification wrapper directly before registering the task:

npm.cmd run verify:scheduled

Each execution stores a UTF-8 log below .local/verification-logs. The directory is ignored by Git, and logs older than 30 days are removed after an execution finishes.

Register the weekday task

Register the task for the current Windows user:

npm.cmd run schedule:install

The default task is named KRI Agent Harness Workspace Verify and runs Monday through Friday at 08:30. It uses one process at a time, retries a failed task up to three times, and runs a missed schedule when the signed-in user becomes available.

The task uses an interactive, limited Windows principal. The computer must be running and the registered user must be signed in before the process can execute.

To use a different time, call the registration script directly with an HH:mm value:

.\scripts\register-scheduled-verification.ps1 -At <HH:mm>

Remove the task

npm.cmd run schedule:uninstall

Removing the task does not remove retained verification logs.

Result handling

Task Scheduler records the verification process exit code:

  • 0: all registered Agent Harness documents are current;
  • non-zero: a missing repository, invalid profile, stale generated document, or another verification error was detected.

Treat a failed task as a drift report, not as authorization for automatic repair:

  1. inspect the newest log in .local/verification-logs;
  2. update the authoritative .agent source when a rule changed;
  3. run plan and obtain approval for affected repositories;
  4. run the narrowest applicable compose command;
  5. run the verification again and commit each repository independently after approval.