The self-test workflow on github.concur.com runs the action with:
- gitea-url=https://gitea.weiker.me (binary download source)
- dry-run=true (avoids PR# mismatch between GHE and Gitea)
This validates:
- Binary download and checksum verification works from a GitHub runner
- GITHUB_SERVER_URL/GITHUB_REPOSITORY env vars are correctly passed
- AiCore provider authenticates and LLM call succeeds
When strat/review-bot has its own releases, remove gitea-url override and dry-run.
Both composite actions hardcoded:
- Gitea's /api/v1/ releases endpoint path
- 'rodin/review-bot' as the action source repo
- .gitea version used GITEA_URL/GITEA_REPO env vars instead of
GITHUB_SERVER_URL/GITHUB_REPOSITORY
Fix:
- Add 'action-repo' input (default: empty) to allow explicit override
- Auto-detect VCS host from server_url: URLs containing 'gitea' use
/api/v1/ (Gitea format), all others use /api/v3/ (GitHub format)
- Set smart default action-repo: 'rodin/review-bot' on Gitea,
'strat/review-bot' on GitHub
- Pass server-url and action-repo as step outputs to avoid re-computing
- Fix .gitea action's 'Run review' env to use GITHUB_SERVER_URL and
GITHUB_REPOSITORY (matching .github version and review-bot binary
env var expectations)
- Add .github/workflows/review.yml for self-testing on github.concur.com
Backward compatible: existing Gitea callers using default inputs continue
to resolve rodin/review-bot via /api/v1/ unchanged.
Closes#120
Rodin