docs: complete the README layout tree for all ten packs

[bradAGI]

The layout tree documented four of the ten packs and carried a note asking readers to browse the directory listing "until this section catches up." This catches it up: claude_skill/, langchain/, crewai/, autogen/, pydantic_ai/, and vercel_ai/ are added and the note is dropped.

Checking the existing entries against the packs turned up six that had gone stale as rules shipped. None of these appeared in the tree:

file	missing from the tree
claude_sdk/agent_safety.yaml	CSDK-121..124 (the 120/130/131 shorthand also hid the real set)
claude_sdk/repo.yaml	CSDK-204
claude_sdk/subagent_safety.yaml	CSDK-112
claude_sdk/tool_definition.yaml	CSDK-017, CSDK-018
openai_sdk/agent_safety.yaml	OAI-112
mcp/tool_definition.yaml	MCP-015..022 — the whole go/csharp/php/rust set

Those are corrected here too. The MCP one seemed worth calling out: multi-language MCP coverage is a genuinely notable capability and the README gave no sign it existed.

Also added a line where the ID convention is stated, noting the two packs that don't follow it — autogen/ numbers its agent rules AG2-001..006 in the tool range, and claude_skill/ groups CSKILL-NNN by topic rather than scope. Both currently read as violations of the documented scheme.

Verification. Not checked by eye — I wrote a script that parses the tree out of the README, expands the A-001..005 range notation, and diffs the claimed IDs against every - id: in every file, asserting both directions plus "no rule file left undocumented":

OK: 85 rule files documented, every rule ID matches the tree
$ trustabl rules validate .
OK: 85 rule pack(s), 206 rule(s) valid under rule schema version 14

Happy to contribute that script as a CI check if you'd want the tree kept honest automatically — it's ~40 lines of stdlib Python and would have caught all six drifts at the commit that introduced them. Say the word and I'll open a separate PR.

Deliberately documents only what's on main today, so it doesn't conflict with the rule PRs I have open.

[jhumel-code]

Thanks @bradAGI, this is careful work. I walked the whole tree against origin/main and everything you added or corrected checks out: the six drifts are real, the MCP-015..022 multi-language set and CSDK-121..124 were genuinely invisible, and nothing in the new tree names a rule that does not exist. Dropping the stale note is the right call.

Two things before I take it. ADK-111 (MCPToolset with no tool_filter) landed in google_adk/agent_safety.yaml after you branched, and that line still reads ADK-101..108, ADK-110. A rebase and a re-run of your checker should pick it up.

The ID convention sentence is not quite right. It is not two packs: CSDK-107 and CSDK-108 are tool scope sitting in the 1NN range, CREW-108 likewise, and ADK-008, VAI-006..009 and VAI-012 sit in the tool range at agent and repo scope. That is fourteen rules across five packs, plus claude_skill/. I would rather the line said the numbering is advisory and that a number of rules predate it, than name two packs and leave the rest reading as compliant. While you are in there, please rewrap that paragraph; the inserted sentence runs well past the file's wrap width.

No engine-side pair is needed for this one, scripts/check-rules-sync.sh deliberately excludes docs from the comparison. And yes, please do open the checker as its own PR.

Happy to take this once ADK-111 and the convention sentence are sorted.