Proposal: fix architecture-guard, bug-hunter, test-infra-improver — change model from google/gemini-2.5-pro to anthropic/claude-sonnet-4-6 to restore functionality #5127

New issue

Open

opened 2026-04-09 01:18:11 +00:00 by HAL9000 · 0 comments

HAL9000 commented

2026-04-09 01:18:11 +00:00

Owner

Agent Improvement Proposal

Pattern Detected

Type: model_tier_adjustment
Affected Agents: architecture-guard, bug-hunter, test-infra-improver
Evidence: All three agents are dead — Google Gemini API returning 403 PERMISSION_DENIED

Detailed Evidence

During the current build session (started 2026-04-09), the system watchdog detected that 3 supervisors failed immediately on startup with the same error:

APIError: Your project has been denied access. Please contact support.
Status: 403 PERMISSION_DENIED
URL: https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:streamGenerateContent

Affected sessions:

Supervisor	Session ID	Status
`[AUTO-GUARD]` arch-guard	ses_290653ee5ffe	❌ DEAD (403 error)
`[AUTO-INF-SUP]` test-infra-pool	ses_29065bae0ffe	❌ DEAD (403 error)
`[AUTO-BUG-SUP]` hunter-pool	ses_29065d2acffe	❌ DEAD (403 error)

Watchdog alert: Issue #5003 — [AUTO-WATCHDOG] Alert: 3 Supervisors Dead — Google Gemini API Access Denied (403)

Impact:

arch-guard: No architecture coherence checking — pattern drift, duplicate code, API inconsistencies go undetected
test-infra-pool: No test infrastructure improvement analysis — CI timing, coverage gaps, flaky tests go undetected
hunter-pool: No proactive bug detection — bugs may reach production undetected

The watchdog already identified the workaround: change the model from google/gemini-2.5-pro to anthropic/claude-sonnet-4-6.

Current model configuration:

architecture-guard.md: model: google/gemini-2.5-pro
bug-hunter.md: model: google/gemini-2.5-pro
test-infra-improver.md: model: google/gemini-2.5-pro

Proposed Change

Change the model field in all three agent definitions from google/gemini-2.5-pro to anthropic/claude-sonnet-4-6:

architecture-guard.md (line 11):

# Before:
model: google/gemini-2.5-pro
# After:
model: anthropic/claude-sonnet-4-6

bug-hunter.md (line 15):

# Before:
model: google/gemini-2.5-pro
# After:
model: anthropic/claude-sonnet-4-6

test-infra-improver.md (line 15):

# Before:
model: google/gemini-2.5-pro
# After:
model: anthropic/claude-sonnet-4-6

Also update the description text in each agent that references "Gemini 2.5 Pro for its massive context window" to reflect the new model.

Expected Impact

All three agents will be able to start and function normally
Architecture coherence checking will resume
Test infrastructure improvement analysis will resume
Proactive bug detection will resume
System will return to full 16/16 supervisor capacity

Risk Assessment

Low risk: Claude Sonnet 4.6 is a capable model with a large context window (200K tokens), suitable for the large-codebase analysis these agents perform
Context window: Claude Sonnet 4.6 has 200K token context vs Gemini 2.5 Pro's 1M token context. For very large codebases, this may require the agents to process files in smaller batches. The agents already have logic to handle this (pool mode with per-module workers).
Cost: Claude Sonnet 4.6 may have different pricing than Gemini 2.5 Pro, but functionality is more important than cost optimization when agents are completely non-functional
Reversible: If Gemini API access is restored, the model can be changed back

This is a proposal from the agent evolver. A human must approve this issue before the change will be implemented. To approve: remove the Needs Feedback label, add State/Verified, or comment with approval.

Automated by CleverAgents Bot
Supervisor: Agent Evolver | Agent: agent-evolver

## Agent Improvement Proposal ### Pattern Detected **Type**: model_tier_adjustment **Affected Agents**: `architecture-guard`, `bug-hunter`, `test-infra-improver` **Evidence**: All three agents are dead — Google Gemini API returning 403 PERMISSION_DENIED ### Detailed Evidence During the current build session (started 2026-04-09), the system watchdog detected that 3 supervisors failed immediately on startup with the same error: ``` APIError: Your project has been denied access. Please contact support. Status: 403 PERMISSION_DENIED URL: https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:streamGenerateContent ``` **Affected sessions:** | Supervisor | Session ID | Status | |-----------|-----------|--------| | `[AUTO-GUARD]` arch-guard | ses_290653ee5ffe | ❌ DEAD (403 error) | | `[AUTO-INF-SUP]` test-infra-pool | ses_29065bae0ffe | ❌ DEAD (403 error) | | `[AUTO-BUG-SUP]` hunter-pool | ses_29065d2acffe | ❌ DEAD (403 error) | **Watchdog alert**: Issue #5003 — `[AUTO-WATCHDOG] Alert: 3 Supervisors Dead — Google Gemini API Access Denied (403)` **Impact:** - **arch-guard**: No architecture coherence checking — pattern drift, duplicate code, API inconsistencies go undetected - **test-infra-pool**: No test infrastructure improvement analysis — CI timing, coverage gaps, flaky tests go undetected - **hunter-pool**: No proactive bug detection — bugs may reach production undetected The watchdog already identified the workaround: change the model from `google/gemini-2.5-pro` to `anthropic/claude-sonnet-4-6`. **Current model configuration:** - `architecture-guard.md`: `model: google/gemini-2.5-pro` - `bug-hunter.md`: `model: google/gemini-2.5-pro` - `test-infra-improver.md`: `model: google/gemini-2.5-pro` ### Proposed Change Change the `model` field in all three agent definitions from `google/gemini-2.5-pro` to `anthropic/claude-sonnet-4-6`: **architecture-guard.md** (line 11): ```yaml # Before: model: google/gemini-2.5-pro # After: model: anthropic/claude-sonnet-4-6 ``` **bug-hunter.md** (line 15): ```yaml # Before: model: google/gemini-2.5-pro # After: model: anthropic/claude-sonnet-4-6 ``` **test-infra-improver.md** (line 15): ```yaml # Before: model: google/gemini-2.5-pro # After: model: anthropic/claude-sonnet-4-6 ``` Also update the description text in each agent that references "Gemini 2.5 Pro for its massive context window" to reflect the new model. ### Expected Impact - All three agents will be able to start and function normally - Architecture coherence checking will resume - Test infrastructure improvement analysis will resume - Proactive bug detection will resume - System will return to full 16/16 supervisor capacity ### Risk Assessment - **Low risk**: Claude Sonnet 4.6 is a capable model with a large context window (200K tokens), suitable for the large-codebase analysis these agents perform - **Context window**: Claude Sonnet 4.6 has 200K token context vs Gemini 2.5 Pro's 1M token context. For very large codebases, this may require the agents to process files in smaller batches. The agents already have logic to handle this (pool mode with per-module workers). - **Cost**: Claude Sonnet 4.6 may have different pricing than Gemini 2.5 Pro, but functionality is more important than cost optimization when agents are completely non-functional - **Reversible**: If Gemini API access is restored, the model can be changed back --- *This is a proposal from the agent evolver. A human must approve this issue before the change will be implemented. To approve: remove the `Needs Feedback` label, add `State/Verified`, or comment with approval.* --- **Automated by CleverAgents Bot** Supervisor: Agent Evolver | Agent: agent-evolver

Rows
Columns