docs: refresh v1 authoring and evaluation skills

[willccbb]

Refresh the environment-authoring and evaluation skills against the current v1 APIs. Correct the environment class and trace metadata names, explain Codex session continuation, and configure the solver runtime in the agentic-judge example to match how the judge inherits its runtime policy.

Also document the default Prime runtime, evaluation upload and server behavior, both retry scopes, and the current resolved-config and attempt-log paths used when resuming a run.

Validation: deterministic tests (pytest tests/ -m 'not e2e' -q), Ruff, all pre-commit checks, and the pre-push type check passed. Model-free checks verified the documented API names, defaults, authoring snippets, and output paths. No live model or sandbox runs were performed.

---

Note

Low Risk
Documentation-only changes to skill markdown; no runtime or application code modified.

Overview
Updates the create-environments and evaluate-environments skill docs so they match current v1 naming and CLI behavior.

Authoring (create-environments): Multi-agent guidance now refers to vf.Env / vf.Env[YourConfig] and trace.agent.name instead of Environment and trace.agent_name. Assistant turn continuation is described as transcript resume vs a stateful harness overriding session() (e.g. Codex via ACPHarness), replacing the older resume() wording.

Evaluation (evaluate-environments): Documents default Prime agent runtimes, --no-push upload behavior, and default environment-server vs --no-serve. The agentic-judge example configures --env.solver.runtime.* (judge follows the solver runtime policy; judge runtime flags are ignored). Retries are split into per-agent (--env.<agent>.retries) vs whole-episode (--env.retries). Run output and resume paths use configs/resolved/eval.json and per-launch logs/attempt_<n>/eval.log with logs/latest.

^{Reviewed by Cursor Bugbot for commit 7d80377. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Refresh v1 authoring and evaluation skills docs

• Updates skills/create-environments/SKILL.md: changes resume contract from overriding resume() to overriding session() for persistent live processes; switches extension type from Environment to vf.Env; updates trace agent identity from trace.agent_name to trace.agent.name
• Updates skills/evaluate-environments/SKILL.md: adds Prime sandbox defaults and runtime selection guidance; splits retries into per-agent (--env.<agent>.retries) and whole-episode (--env.retries) scopes; revises output layout to use configs/resolved/eval.json, per-attempt log directories, and a logs/latest symlink; changes agentic-judge example to use solver runtime policy and ignore judge runtime settings
• Risk: these are docs-only changes, but reviewers should confirm the updated session(), vf.Env, trace.agent.name, retry flags, and output paths match current tooling behavior

^{Macroscope summarized 7d80377.}