Map Gemini image-safety finish reasons to content_filter

[PratikDhanave (PratikDhanave)]

Problem

toFinishReason (provider/geminiprovider/agent.go) maps the text content-policy finish reasons — SAFETY, RECITATION, BLOCKLIST, PROHIBITED_CONTENT, SPII — to the canonical content_filter. But the genai SDK also defines image-generation equivalents in the same content-policy category: FinishReasonImageSafety (IMAGE_SAFETY), FinishReasonImageProhibitedContent (IMAGE_PROHIBITED_CONTENT), and FinishReasonImageRecitation (IMAGE_RECITATION). These fall through to the default and return an empty string.

So a response blocked by image-safety policy reports no finish reason, whereas the equivalent text-safety block reports content_filter — an inconsistency within the same semantic category (and a gap vs. the cross-provider content_filter contract).

Fix

Map the three image-safety reasons to content_filter alongside their text counterparts.

Test

Extends the existing TestFinishReason_NonStreaming table with IMAGE_SAFETY, IMAGE_PROHIBITED_CONTENT, and IMAGE_RECITATION rows, each expecting content_filter. They fail before the fix (empty string), pass after.

[Copilot]

🟢 Approval recommended

The mapping change is straightforward, consistent with existing semantics, and is covered by an expanded unit test.

Pull request overview

This PR aligns the Gemini provider’s finish-reason mapping with the framework’s cross-provider content_filter contract by including the genai SDK’s image-safety finish reasons, preventing blocked image-generation responses from reporting an empty finish reason.

Changes:

• Map IMAGE_SAFETY, IMAGE_PROHIBITED_CONTENT, and IMAGE_RECITATION to the canonical content_filter finish reason.
• Extend the existing non-streaming finish-reason table test to cover the new image-safety reasons.

File summaries

File	Description
provider/geminiprovider/agent.go	Extends toFinishReason to treat image-safety finish reasons as content_filter.
provider/geminiprovider/agent_test.go	Adds table-driven test cases asserting content_filter for image-safety finish reasons.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

Scope: user-visible behavior (unexported toFinishReason helper, but the returned content_filter value is observed by callers via the response's finish-reason field)

Changed Go contract: toFinishReason in provider/geminiprovider/agent.go now maps genai.FinishReasonImageSafety, genai.FinishReasonImageProhibitedContent, and genai.FinishReasonImageRecitation to "content_filter", alongside the existing text-safety reasons (SAFETY, RECITATION, BLOCKLIST, PROHIBITED_CONTENT, SPII).

Upstream evidence reviewed: python/packages/gemini/agent_framework_gemini/_chat_client.py, _FINISH_REASON_MAP (lines ~315–326). Python maps SAFETY, RECITATION, LANGUAGE, BLOCKLIST, PROHIBITED_CONTENT, SPII, IMAGE_SAFETY, IMAGE_PROHIBITED_CONTENT, and IMAGE_RECITATION all to "content_filter".

Result: aligned. The three image-safety reasons added in this PR match upstream Python's mapping to content_filter exactly, closing the parity gap described in the PR. No exported Go API surface changed (only internal mapping logic + tests), so this is not a public API change.

Note (non-blocking, out of scope for this PR): upstream Python also maps LANGUAGE → content_filter, which Go's toFinishReason does not currently handle. This is a pre-existing gap unrelated to this PR's fix and may be worth a follow-up.

Warning

Firewall blocked 2 domains

The following domains were blocked by the firewall during workflow execution:

• github.com
• raw.githubusercontent.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "github.com"
    - "raw.githubusercontent.com"

See Network Configuration for more information.

Generated by Go API Consistency Review Agent · copilot · auto · 36.4 AIC · ⌖ 5.37 AIC · ⊞ 9.5K · ◷