[CI] Also disable prompt caching for NVIDIA inference proxy

NVIDIA/OSMO's claude workflow (testbot-respond.yaml) sets BOTH
CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1 and DISABLE_PROMPT_CACHING=1
because the NVIDIA gateway rejects two fields the SDK sends by
default: `context_management` (beta) and `cache_control.ephemeral.scope`
(prompt caching). Adding the second env var preemptively so we don't
hit a follow-up 400 after the first fix lands.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>

Author: kevalmorabia97

.github/workflows/claude.yml

@@ -68,11 +68,13 @@ jobs:
         uses: anthropics/claude-code-action@v1
         env:
           ANTHROPIC_BASE_URL: ${{ secrets.ANTHROPIC_BASE_URL }}
-          # The NVIDIA inference proxy (LiteLLM-based) rejects the
-          # `context_management` field that Claude Code's SDK sends by
-          # default. Disabling experimental betas drops this field so
-          # the proxy accepts the request.
+          # NVIDIA inference proxy (LiteLLM-based) rejects two fields
+          # the Claude Code SDK sends by default. Set per NVIDIA/OSMO's
+          # workflow which has hit and solved both issues:
+          #   - `context_management` → disable experimental betas
+          #   - `cache_control.ephemeral.scope` → disable prompt caching
           CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS: "1"
+          DISABLE_PROMPT_CACHING: "1"
         with:
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           claude_args: |

.github/workflows/claude_review.yml

@@ -60,11 +60,13 @@ jobs:
         uses: anthropics/claude-code-action@v1
         env:
           ANTHROPIC_BASE_URL: ${{ secrets.ANTHROPIC_BASE_URL }}
-          # The NVIDIA inference proxy (LiteLLM-based) rejects the
-          # `context_management` field that Claude Code's SDK sends by
-          # default. Disabling experimental betas drops this field so
-          # the proxy accepts the request.
+          # NVIDIA inference proxy (LiteLLM-based) rejects two fields
+          # the Claude Code SDK sends by default. Set per NVIDIA/OSMO's
+          # workflow which has hit and solved both issues:
+          #   - `context_management` → disable experimental betas
+          #   - `cache_control.ephemeral.scope` → disable prompt caching
           CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS: "1"
+          DISABLE_PROMPT_CACHING: "1"
         with:
           anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
           trigger_phrase: "/claude review"