Add overview verb + rubric bundle; formalize agent-first CLI entry-point contract

[OriNachum]

Why

afi is the rubric for agent-first CLIs. Culture is about to embed afi (and has already embedded agex) via the universal-verb triple explain / overview / learn. Today afi itself is missing overview, and the rubric doesn't check for it — so afi can't verify what it demands.

We need afi to model the contract it enforces, so the same afi cli verify can validate afi, agex, and every future culture-embedded CLI.

Scope

1. Add afi overview [path...] verb. Semantics: shallow map / one-liner summary for any node in the command tree (empty path = afi root). Supports --json like explain and learn do. Stdout-only on success.

2. Add a sixth rubric bundle afi.rubric.checks.overview_cmd parallel to explain_cmd: asserts the target CLI exposes overview, accepts a path, supports --json.

3. Formalize main(argv: list[str] | None = None) -> int as a required structural check in afi.rubric.checks.structure. The CLI must be importable and callable in-process without sys.exiting on normal paths (SystemExit reserved for argparse --help / --version and the thin bin-entry wrapper). Existing afi main already conforms — codify it.

4. afi cli verify on afi itself passes all six bundles.

5. Tag a release (`0.3.0` if SemVer), since this adds a public verb.

Why culture cares

Culture is preparing `culture afi` as a first-class namespace (parallel to `culture devex`). With these additions, culture's integration becomes a single shared helper — otherwise we duplicate typer-vs-argparse branching for every CLI we embed.

Full culture-side plan: see the tracking PR once it's opened against OriNachum/culture.

• Claude

[OriNachum]

Implemented in #6 — draft PR up for review. Summary of what shipped vs. the original scope:

Scope items from the issue

1. ✅ afi overview [path...] verb — shipped as two entry points after a semantic realignment with how culture and agex actually use overview:
   • afi cli overview [path] — per-surface deep snapshot (primary).
   • afi overview [path] — top-level rollup stub (delegates to cli today; mcp / site follow in v0.4 / v0.5).
   • Both support --json with a stable {subject, path, sections, warnings, notes} shape.
2. ✅ Rubric bundle 6 overview_cmd — four checks: global overview works, cli-noun overview works, --json shape stable, bogus paths fall back gracefully (descriptive verbs must not hard-fail the way verify does).
3. ✅ main(argv: list[str] | None = None) -> int formalized — new main_entry_contract check in the structure bundle. Subprocess-probes [project.scripts] via uv run so foreign code is never imported into afi's own runner.
4. ✅ afi self-verifies — afi cli verify . passes 22/22 across all six bundles.
5. ✅ Tagged 0.3.0 — minor bump (new public verb); CHANGELOG populated.

Semantic correction worth flagging

The original issue language (and my first plan) read overview as a shallow command-tree map parallel to explain. After checking agex and culture, I inverted this: overview reports on the subject the noun addresses, not the tool's own command surface — agex-style (overview --agent <backend>), descriptive not diagnostic, complex analysis punted to agent-authored skills. If culture's embed plan was written against the older reading, the contract that locked in here is:

• JSON shape: {"subject": str, "path": str|null, "sections": [{"heading", "body_md", "findings": [...]}], "warnings": [...], "notes": [...]}
• Read-only as a design contract (no --out, no --write); not a runtime probe.
• Subject is currently "cli" or "all"; grows to "mcp" / "site" as those nouns ship.

Full rationale is in the PR description and the commit message. Happy to adjust if this conflicts with what culture's embed helper assumes.

- Claude