Wire public mock env task runtime

[bingran-you]

Summary

• point the public task manifest at the public env0 base image and mock-* service CLIs
• add BenchFlow environment manifests to example tasks so bench eval run can start mock services
• update imported task Dockerfiles, stale seed tests, and contributor validation docs

Validation

• BENCHFLOW_REWARD_LENIENT=1 bench eval run --tasks-dir example_tasks --include gdrive-archive-stale-drafts --agent oracle --sandbox docker --jobs-dir .local/bf-jobs-public-example-smoke --context-root . --concurrency 1 --build-concurrency 1
• BENCHFLOW_REWARD_LENIENT=1 bench eval run --tasks-dir example_tasks --agent oracle --sandbox docker --jobs-dir .local/bf-jobs-public-examples-oracle --context-root . --concurrency 1 --build-concurrency 1
• built all tasks/*/environment/Dockerfile images against ghcr.io/benchflow-ai/env0:0.1.0
• scripts/smoke_dev.sh
• PULL_BASE=0 scripts/smoke_docker_examples.sh
• bench tasks check over all example_tasks/* and tasks/*
• cd packages/environments/mock-gdoc && uv run --extra dev pytest tests -q
• cd packages/environments/mock-gcal && uv run --extra dev pytest tests -q
• git diff --check

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bec6860153

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Hide the baked task payload from agents

When these tasks/* Dockerfiles are evaluated on the public env0 image, the seed payload is still copied under /tasks and exposed through TASKS_DIR; Docker COPY leaves those files world-readable, so the normal agent user can inspect data/needles.py and recover target IDs/answers before acting. The same pattern appears across the updated imported task Dockerfiles; move the payload under /var/lib/task with root-only permissions, or otherwise remove agent read access after seeding.

Useful? React with 👍 / 👎.