fix(fullsend): switch back to opus

[gashcrumb]

This change reverts the model switch to Sonnet but keeps the longer timeout in place for reviews

rh-pre-commit.version: 2.4.0
rh-pre-commit.check-secrets: ENABLED

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[fullsend-ai-review]

🤖 Finished Review · ❌ Failure (validation failed after 1 iteration(s)) · Started 11:00 AM UTC · Completed 11:02 AM UTC

Commit: cd08061 · View workflow run →

Runtime: claude · Model: claude-sonnet-4-5@20250929 · Effort: high

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 12:03 PM UTC · Completed 12:14 PM UTC

Commit: cd08061 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $3.74

[fullsend-ai-retro]

Retro: PR #192 — Cascading agent failure resolved by human intervention

PR #192 resolved a cascading agent failure that spanned ~4 hours and cost ~$14.53 in agent compute with zero successful agent-generated fixes.

What happened

1. PR chore: upgrade fullsend review harness #181 (merged Sep 9 08:11 UTC): Added model: sonnet to the review harness. The alias resolved to claude-sonnet-4-6, which is not provisioned on Vertex AI — all reviews broke.
2. Retro → Issue Review harness model sonnet resolves to unavailable claude-sonnet-4-6 on Vertex AI #187 (08:25 UTC): Detected the breakage but recommended claude-sonnet-4-5@20250929, which is also unavailable on Vertex AI.
3. Code agent → PR fix: pin review harness model to available Vertex AI version #188 (08:35 UTC): Implemented the bad recommendation with a non-conforming PR title (# in scope). Human fixed the title and merged at 10:14 UTC. Reviews still broken.
4. Human → PR fix(fullsend): use the correct model identifier #191 (10:47 UTC): Tried the correct model (claude-sonnet-4@20250514). Closed at 10:54 UTC — chicken-and-egg deadlock prevented review.
5. Retro → Issue Review harness model claude-sonnet-4-5@20250929 is unavailable — all reviews blocked #193 (11:06 UTC): Filed correct recommendation, but PR fix(fullsend): switch back to opus #192 was already open with a different approach.
6. Code agent → PR fix(#193): update review harness model to available Vertex AI version #194 (11:16 UTC): Correct model but repeated the bad PR title pattern. Review also failed.
7. Human → PR fix(fullsend): switch back to opus #192 (10:59 UTC, merged 12:02 UTC): Removed the model override entirely, reverting to Opus. Problem resolved.

Evidence for existing issues

• Dispatch config validation creates unrecoverable deadlock when main config is invalid fullsend-ai/fullsend#6186 (config deadlock): This incident adds another concrete example — 4 fix PRs (fix: pin review harness model to available Vertex AI version #188, fix(fullsend): use the correct model identifier #191, fix(fullsend): switch back to opus #192, fix(#193): update review harness model to available Vertex AI version #194) could not be reviewed because the review agent reads model config from the broken base branch.
• Validate model availability before review sub-agent dispatch to avoid wasted retries fullsend-ai/fullsend#6964, #6922 (model availability): The retro agent trusted an error message's model suggestion without validating availability, propagating a bad recommendation that cost ~$1.50 in wasted code agent and triage runs.
• Document PR title scope character restrictions in AGENTS.md #189 (PR title scope): Code agent repeated the #-in-scope mistake on PR fix(#193): update review harness model to available Vertex AI version #194, confirming the AGENTS.md update is still needed.

Immediate concern

Issue #193 and PR #194 remain open with ready-to-code labels despite being superseded by PR #192. The issue proposes switching to a different Sonnet model, which conflicts with the human's deliberate decision to revert to Opus. See proposal below.

Proposals filed

• Close superseded Issue #193 and PR #194 to prevent wasted agent cycles (in redhat-developer/rhdh-cli)

Proposals skipped (target repo not allowed)

File manually or update create_issues.allow_targets in config.yaml:

• Retro agent should detect and flag agent-filed issues superseded by merged PRs (fullsend-ai/agents)