feat(agents): add GPT-5.6 to Pi model catalogs

[mmabrouk]

Summary

• Add the concrete GPT-5.6 sol, terra, and luna models to the OpenAI model list.
• Expose the openai-codex provider and GPT-5.6 models through the Pi harness capability document.
• Align the runner dependency and compose environment examples with the pinned Pi version, 0.80.6.
• Default Pi installation to enabled in the five non-snapshot GitHub compose examples.

Context

PR #5046 now owns the Daytona snapshot and Pi installation behavior. This PR is rebased directly on that merged change and does not duplicate or override its snapshot logic.

The resulting contract is:

• non-snapshot GitHub compose files default AGENTA_AGENT_SANDBOX_PI_INSTALLED=true, so Pi is installed in each sandbox by default;
• the dev compose keeps DAYTONA_SNAPSHOT=agenta-sandbox-pi and AGENTA_AGENT_SANDBOX_PI_INSTALLED=false, because the selected snapshot already contains the pinned Pi version;
• AGENTA_AGENT_SANDBOX_PI_VERSION remains an optional override, documented as 0.80.6;
• model catalogs use the concrete GPT-5.6 variant IDs only. The unsupported bare gpt-5.6 alias is not advertised.

Validation

• pnpm install --offline --frozen-lockfile
• Pi CLI version: 0.80.6
• runner typecheck
• runner extension build
• Daytona unit tests: 7 passed
• runner unit tests excluding the known stale replay capture: 739 passed
• invoke-handler tests: 45 passed
• targeted SDK capability/model tests: 4 passed
• Ruff format and lint checks
• live Daytona snapshot smoke: Pi 0.80.6, Claude/Codex/OpenCode/geesefs present, and both openai and openai-codex return the three GPT-5.6 variants
• active Daytona sandbox count returned to zero after the smoke test

The broader SDK test selection reports one unchanged baseline failure, test_unknown_harness_is_permissive, which is also failing on big-agents.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agenta-documentation	Ready	Preview, Comment	Jul 10, 2026 12:50pm

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 45f59c74-dc85-4758-a1f7-968cc691e724

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/pi-gpt-5-6-models

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[mmabrouk]

🤖 The AI agent says:

The PR now points to 99b63d07389ef77e83ea77184549c4be33e5f021, including the compose default fix.

Validation completed:

• styling and contribution checks passed;
• runner unit, integration, and acceptance checks passed;
• Vercel and CodeRabbit checks passed;
• API, SDK, and web unit jobs reproduce the same baseline failures as big-agents (6 API, 2 SDK, 22 web failures).

I attempted to enable auto-merge twice after all checks completed. GitHub rejected it with Pull request is in unstable status because of those failed baseline jobs. No merge bypass was used. The PR remains open and ready for queueing once the baseline checks are repaired or the repository policy allows those known failures.

[mmabrouk]

lgtm

[mmabrouk]

🤖 The AI agent says:

Thanks for the LGTM. I retried enabling auto-merge on PR #5198 after the review landed, but GitHub still rejects it with Pull request is in unstable status.

The remaining failures are the unchanged big-agents baseline jobs:

• API unit tests: 6 failures
• SDK unit tests: 2 failures
• web unit tests: 22 failures

The runner, styling, contribution, preview, and CodeRabbit checks remain green. The PR is not in the merge queue because GitHub will not create the auto-merge request while these failed checks remain.