fix(targets): add Pi SDK OpenAI target

[christso]

Summary

• Repoint the existing pi-sdk target to the standard OpenAI-compatible endpoint with gpt-5.5 and thinking: low.
• Add a named pi-sdk-openai target with the same OpenAI-compatible Pi SDK config.
• Set gpt-5.5, thinking: low, and stream_log: raw on Pi SDK targets touched in this PR.
• Remove explicit tools allowlists from Pi SDK targets so Pi uses its default tool set.
• Replace tracked target/template log_format: json usages with stream_log: raw.
• Keep deprecated log_format compatibility in code, but allow stream_log in validation for providers that already consume it.

Verification

• Live Pi SDK sample evals:

  NODE_PATH=/home/entity/.local/lib/node_modules bun apps/cli/src/cli.ts eval examples/features/basic/evals/dataset.eval.yaml \
    --targets .agentv/targets.yaml --target pi-sdk --test-id shorthand-string-example \
    --output .agentv/verification/age-33/basic-shorthand-pi-sdk-live-gpt55

  • PASS 1/1, target pi-sdk, provider log written.

  NODE_PATH=/home/entity/.local/lib/node_modules bun apps/cli/src/cli.ts eval examples/features/basic/evals/dataset.eval.yaml \
    --targets .agentv/targets.yaml --target pi-sdk-openai --test-id shorthand-string-example \
    --output .agentv/verification/age-33/basic-shorthand-pi-sdk-openai-live-gpt55

  • PASS 1/1, target pi-sdk-openai, provider log written.
• Export-screening Pi SDK execution smoke:

  NODE_PATH=/home/entity/.local/lib/node_modules bun apps/cli/src/cli.ts eval examples/showcase/export-screening/evals/dataset.eval.yaml \
    --targets .agentv/targets.yaml --target pi-sdk --test-id exp-high-001 \
    --output .agentv/verification/age-33/export-exp-high-001-pi-sdk-live-gpt55

  • Target executed successfully and final assistant message contained valid JSON with riskLevel: High.
  • The strict code-grader failed because agent transcript/tool output is included before the final JSON in response.md; this is an eval/grader compatibility issue for agent targets, not target execution failure.
• Deprecated-log-format warning check:

  bun --filter @agentv/core build
  bun apps/cli/src/cli.ts eval examples/features/basic/evals/dataset.eval.yaml \
    --targets .agentv/targets.yaml --target copilot-sdk --test-id shorthand-string-example --dry-run \
    --output .agentv/verification/age-33/basic-shorthand-copilot-sdk-stream-log-dry-run-fixed-rebuilt

  • PASS 1/1 dry-run.
  • Confirmed no log_format deprecation warning and no Unknown setting 'stream_log' warning.
• Pre-push hook passed:
  • bun --filter @agentv/core typecheck
  • bun --filter @agentv/phoenix-adapter typecheck
  • bun --filter agentv typecheck
  • biome check .

Notes

For strict JSON classification evals, direct LLM targets are still a better fit than agent targets because agent target artifacts may include tool transcripts in addition to the final assistant answer.

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	07ae6f7
Status:	✅  Deploy successful!
Preview URL:	https://6d85e070.agentv.pages.dev
Branch Preview URL:	https://fix-pi-sdk-openai-target.agentv.pages.dev

View logs