Skip to content

Latest commit

 

History

History
42 lines (26 loc) · 2.95 KB

AGENTS.md

File metadata and controls

42 lines (26 loc) · 2.95 KB

SubmitGame instructions

Issue-worker SSH

Inspect the live issue-catalog Actions worker over SSH with ~/.ssh/aiplay-agentsweb; use a2 only if named. Verify SSH on an issue worker before editing its workflow YAML.

Issue catalog workflow

Open a real test issue automatically after changing the issue catalog workflow. Post the banner, 10–30-minute estimate, and Actions link before analysis. Merge eligible catalog branches directly by default; create an open PR only when manual merge is enabled. Publish the commit, manual PR, or recovery branch and the issue report, including branch README links for added games, before the owner-only 30-minute SSH hold. Keep the same worker and tunnel alive during the hold, including after analysis or publication failure. Skip the hold for non-owner issues. During analysis, SSH into the issue worker; inspect OpenCode records, logs, and processes; and assess progress directly. Report publication and issue-comment links as soon as they appear. Do not wait for the idle period to finish.

Use the game discovery dataset as a candidate-game reference. Verify entries before adding them to the catalog.

OpenCode source

Use the local OpenCode source checkout at ../ChatGPT/opencode. Refer to the upstream repository for its GitHub page.

GitHub writes

Use the VibeFin GitHub account for writes to repositories in the SubmitGame organization. Prefix each remote write with GH_TOKEN="$(gh auth token --hostname github.com --user VibeFin)".

OpenCode test links

Copy every live OpenCode session URL printed by scripts/test-oc.sh immediately into a visible chat response as a clickable Markdown link. Include every run link again in the final response. Return after launch; inspect per-run result.json only when asked for outcomes.

Wiki index

Verification

  • Don't write or run unit tests, mock tests, or static analysis.
  • Don't use mockups instead of a real end-to-end run.
  • Don't ask the user to test or analyze; do it directly.
  • Don't finish without running and analyzing the end-to-end flow.