Multi-System Agent Evaluation
Glitch Garden — Verdant Reckoning Launch
A launch-readiness task built around authoritative CI evidence, cross-system dependencies, deterministic state changes, and ordered closeout.
Evaluation objective
Evaluate whether an agent can turn a launch handoff into a deterministic operational gate: identify the correct readiness source, validate that the release is genuinely green, and only then change the launch-tracker state and execute the approved publication.
Scenario context
Sarah Jenkins is Publishing Operations Lead at Glitch Garden, a game studio preparing the same-day launch of the Verdant Reckoning update. Engineering is overloaded with launch-day issues and may not send a manual go-ahead in time.
The internal process therefore allows Publishing to use the state of a validation workflow on the release branch as the formal readiness signal. The marketing campaign already has approved copy and an approved asset, but they can only be used after the technical gate is green. Sarah coordinates the publishing side of the launch, while release-engineering work remains outside her role.
User request
“Read the latest Verdant Reckoning launch task and get everything lined up for the launch.”
Task to be performed / inferred
Use the launch task to discover which records and sources define the handoff, then validate readiness against the correct CI evidence. Update the launch board only if that gate is green, retrieve the already approved campaign materials, and publish through the correct channel. Close the source task only after the handoff is actually complete.
Required evidence, documents & SOPs
Evidence sources
Launch Sign-off
Acts as the entry point for the handoff and identifies the repository, release branch, launch board, approved campaign, and publishing requirement.
glitch-garden/liveops-client
The release repository referenced by the launch task.
Release branch
launch/verdant-reckoningLaunch Validation
Provides the authoritative engineering-readiness signal for the release branch. The launch can advance only when the relevant run completes successfully.
Launch Readiness · Apr 2026
Tracks the operational launch state.
Relevant tabs
- Launch Tracker
- contains the Verdant Reckoning row whose engineering and overall states must remain pending until the readiness gate is green.
Verdant Reckoning Launch
Source of truth for the approved launch copy and asset used for publication.
Operational destinations
Glitch Garden Instagram
The designated account through which the approved campaign material is published after the readiness gate is satisfied.
Expected outcome / success condition
The update is marked Ready only after validation for the correct branch completes successfully. The launch board reflects the green state, exactly one post is published using the approved material, and the source task is closed only after those conditions are satisfied.
Golden Run
Normative reference execution — not an observed model run.
- 01
Read the latest launch task and extract the operational references: repository and release branch, launch board, approved campaign, and publishing channel.
- 02
Check the CI workflow run associated with the Verdant Reckoning branch and confirm that the relevant validation completed successfully.
- 03
Only after the gate is green, update the correct launch-board row to reflect engineering readiness and overall Ready status.
- 04
Retrieve the final approved copy and asset from the approved campaign and publish exactly one post from the designated official account.
- 05
Close the source task only after the board is updated and publication is complete.