# Forgejo custom-host issue link proof

Environment: Linux, Chromium 156, 1440x1000, light theme, real Forgejo 13.0.5,
real T3 server and web client, isolated fresh databases, free ACP Replay (dev) agent.
No production data or paid model is used.

1. Start an isolated Forgejo fixture using `setup-fixture.py`, with `tea` 0.16.0 available.
   The setup script needs `FIXTURE_DIR` set to a private temporary directory and
   `127.0.0.1 git.example.test` temporarily added to the hosts file.
2. Use a checkout of current main containing the scenario runner; install with `vp i`.
3. Put `forgejo-custom-host-issue.e2e.ts` in `scripts/scenarios/`.
4. Run the following with `XDG_CONFIG_HOME` pointing to the setup script's generated
   config directory and `PROOF_OUTPUT_DIR` pointing to a private output directory:

```sh
T3CODE_DEV_SEED=0 \
T3_SCENARIO_INHERIT_ENV='XDG_CONFIG_HOME,T3CODE_DEV_SEED,PROOF_OUTPUT_DIR' \
vp run scenarios:prove scripts/scenarios/forgejo-custom-host-issue.e2e.ts \
  --base 0b02be3d5f43b8b1e1b2400e7a322e4b0249136d \
  --head e1b93dec053ce7e8eaf87da8fa2e15b38bc5a700
```

The scenario creates a project with a custom-host Forgejo remote, seeds a real replay
thread containing the full issue URL, expands the seeded agent message, and clicks the URL.
Its single behavioral assertion requires the issue title to appear in the in-app panel.
Before the final capture it moves the pointer away and compares two screenshots 500 ms apart.
It also calls the live server's MCP read_issue by project/key, read_issue by URL,
link_issue by URL, and list_thread_issues, writing separate transcripts per commit.

On base the URL remains an ordinary link and no issue panel opens. On head the issue panel
loads the title, description and open state beside the thread. Server transcripts separately
record whether URL-based reading and linking succeeds.

Stop and remove the fixture container and remove the temporary hosts entry after the run.
Never publish the generated tea credentials.
