docs: make Insight Codex workflows discoverable and verifiable

[cgfixit]

Proposed changes

Repository workflows existed only under .codex/skills, outside documented Codex
repository discovery, and two workflows declared name: optimize. Add four uniquely
named insight-* entrypoints under .agents/skills that reuse the maintained workflows.
Correct the optimizer name/code fence, lazy state-load verification, offline preflight,
and CI smoke/integration guidance. Link a source-backed setup review from the README.

Invariant/governance impact: documentation and skill discovery only; runtime, dependency
manifests, CI gates, output formats, and model loading behavior are unchanged.

Type of change

• Documentation / Codex workflow correction
• Runtime behavior change
• Dependency change

Benefits / why

New tasks can discover repository-specific workflows without generic personal-skill
name collisions. Verification guidance follows actual state restoration and CI behavior.
The review records concrete tokenizer, integration-fixture, state-validation, and
documentation follow-ups for future focused changes.

Risks to monitor

Restart/open a new task if the current skill catalog does not refresh. Optional E2E
tests already fail because AutoTokenizer is not a runtime module attribute; this
documentation-only PR records the failure rather than changing test/runtime behavior.

Completed checklist

• Constrained editable development installation in an isolated local .venv
• pip check: no broken requirements
• Ruff lint and format pass
• Strict mypy passes (10 source files)
• Offline unit tests: 118 passed
• CI production CLI smoke body passes with fake model/tokenizer boundaries
• Four skill metadata checks and workflow links pass
• Lightweight import resolves to this checkout without importing torch
• Diff whitespace and named-file staging reviewed
• Optional E2E suite passes: pre-existing fixture AttributeError reproduced;
  stopped at first failure
• Python 3.13 local environment / real BERT inference: not run for docs-only scope

Further comments

Local tests used the available Windows Python 3.12.0rc3. GitHub CI supplies stable
Python 3.12/3.13 coverage. No model weights or generated outputs are committed.
Suggested next skills: insight-tokenizer-check and insight-state-contract.

Branch naming / merge order / base commit

Branch: codex/insight-codex-setup, created from origin/main.
Base commit: 8097406 (fix: make lazy-import AttributeError test pass ruff B009/B018).
Target: main. Standalone draft PR; no merge dependencies.

[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 2026-09-06T15:02:14.408153Z	d54a281	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.