fix(cli): finish policy-slot initialization without a traceback

[Birfy]

Reproduced problem

agentdescent init selection --kind policy_slot writes its starter file and then exits with KeyError: 'path' on current main. Policy-slot specs use data.problems, but cmd_init assumes every kind uses a row-data path.

Fix

• Branch the next-step guidance for policy slots and explain the zero-argument problem builder and import allowlist.
• Clarify init --data help and the matching documentation.
• Preserve generated specs and the existing import permissions.
• Add default/custom-ref regression cases and exercise the actual installed console command in the clean-wheel smoke gate.

This patch changes only CLI onboarding, tests, documentation and the changelog. It does not touch the engine work in the other open PRs.

Validation

• 44 focused CLI, policy-slot, run-store and wheel-smoke tests pass with provider credentials unset.
• Wheel and sdist build successfully; twine check dist/* passes.
• Clean installed-wheel CLI/demo/MCP smoke passes on Python 3.12 and 3.13. Clean sdist install/runtime smoke also passes.
• mkdocs build --clean --strict passes after the cold font cache was populated (4.20 seconds on the successful run).
• Independent review caught and corrected the initial mapping-versus-builder wording; the final diff was reviewed clean.
• Full local comparison: unchanged main has 2,983 passed / 56 skipped / 19 failed; the branch has 2,985 passed / 56 skipped / the identical 19 failures. There are no new failing test names. Eighteen sandbox-dependent examples cannot start nested Bubblewrap in this cloud runtime (NETLINK_ROUTE socket: Operation not permitted); the remaining failure is the installed Chromium PDF renderer. The same sandbox limitation persists through the supported expanded execution mode; no isolation or security settings were changed. These local limitations are not claimed as passing checks.
• Final published-head focused checks pass again: 44 tests. Exact-head hosted CI is green at b8b99dde4f4a2c80ded5eabb3e59bcbc0dff972a: full tests, built wheel and installed-wheel smoke on Python 3.9, 3.10, 3.11 and 3.12, plus the strict docs build. The PR docs deploy is skipped as expected.

Independent review and exact-head CI are green. This PR is ready for review. No release or merge is included.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-10T01:33:08.511534Z	b8b99dd	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.