[durandom]

Summary

• Pass --community-registry so the todo catalog index resolves plugin images from ghcr.io/rhdh-parasol/... instead of the hardcoded upstream default.
• Remove the GHCR visibility PATCH experiment workflow (the unofficial API returns 404; packages were made public in the UI).

Test plan

• Generate Catalog Index on main logs ghcr.io/rhdh-parasol/rhdh-plugin-export-overlays for the two todo packages
• plugin-catalog-index:latest still pulls anonymously

Made with Cursor

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 12:03 PM UTC · Completed 12:07 PM UTC

Commit: 4a8bf08 · View workflow run →

Runtime: claude · Model: claude-opus-4-6 · Cost: $0.67

[fullsend-ai-review]

Review skipped — this PR is already merged.

The /fs-review command only reviews open pull requests.

_{Posted by fullsend pre-review check}

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 12:03 PM UTC · Completed 12:22 PM UTC

Commit: 4a8bf08 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $5.00

[fullsend-ai-review]

Review skipped — this PR is already merged.

The /fs-review command only reviews open pull requests.

_{Posted by fullsend post-review check}

[fullsend-ai-retro]

Retro: PR #15 — ci: point community catalog at fork GHCR

PR #15 was a trivial 2-file follow-up fix (+1/−107) by the repo owner (durandom) that added the --community-registry flag to the catalog index workflow and deleted a non-functional experimental workflow. It was merged 12 seconds after creation. PR #16, another 1-file follow-up fix, was merged 6 seconds after creation. Both were part of a rapid-iteration session to finish the fork's GHCR publish pipeline (issue #5).

What happened

• The review agent was dispatched on PR ci: point community catalog at fork GHCR #15's opened event (12:02:19 UTC), but the PR was already merged by 12:02:28 UTC.
• The pre-review script detected the merged state and posted a skip comment, but exited 0 — allowing the agent to proceed.
• The agent ran for ~130 seconds in a sandbox (62.7s setup), produced an APPROVE verdict with zero findings, spent $0.67 in tokens, and the post-review script discarded all output.
• PR fix: forward community registry into catalog bootstrap #16 followed the same pattern: merged before the review bot could act, review skipped.
• Neither PR received any review coverage — agent or human.

Existing issues that cover these findings

• fullsend-ai/agents#74 — Post-review should deliver findings as informational comments on merged PRs instead of discarding them. PR ci: point community catalog at fork GHCR #15 is another instance: $0.67 spent on a review whose output was silently dropped.
• fullsend-ai/fullsend#4768 — Review agent should perform post-merge informational review when no pre-merge review occurred. PRs ci: point community catalog at fork GHCR #15 and fix: forward community registry into catalog bootstrap #16 both had zero pre-merge reviews and contained CI workflow changes.
• fullsend-ai/fullsend#5399 / #5817 — Skip retro dispatch for merged PRs with zero meaningful agent involvement. PR ci: point community catalog at fork GHCR #15 had no successful review, no code agent, and a 2-file trivial change — retro dispatch adds cost without novel findings.
• fullsend-ai/agents#1116 — Review sub-agent model config should use deployment-agnostic aliases. On PR ci: publish only todo to fork GHCR until the pipeline is proven #10, review run 5 (Sep 3) used claude-opus-4-8 despite the repo config specifying claude-opus-4-6, causing a policy_denied failure — the last commit was never reviewed.

Assessment

The workflow went as well as it could given the structural constraint: the repo owner merged PRs faster than the review bot could start. This is a legitimate workflow pattern for a fork owner iterating on CI/CD infrastructure. All improvement opportunities are already tracked in existing open issues. No new proposals are warranted.