docs(hook-observability): record the three output-cap deltas from the 2026-09-04 hooks-reference audit

[kyle-sexton]

Summary

The 2026-09-04 doc-alignment audit (raw fetch of https://code.claude.com/docs/en/hooks.md, every claim checked against source) confirmed docs/conventions/hook-observability/README.md accurate on its channel claims but found three deltas the convention does not carry:

1. Hook output over 10,000 characters now overflows to a file with a preview and the path, rather than being truncated. The repo's docs say "cap" only.
2. The additionalContext cap is 10,000 characters per value, with no shared pool across hooks ("Claude receives all of the values").
3. A separate 2,000-character cap, shared across every hook responding to one call, governs the auto-mode classifier note (classifierContext), not additionalContext. That section also states the note is ignored for background hooks and discarded on read-only lookups.

The third one matters because a peer session misread it and filed typos-format's 8,000-character self-cap as a bug; the finding was withdrawn, but the convention doc is where the next reader should have found the answer.

Fix

Three short additions to docs/conventions/hook-observability/README.md under the additionalContext and output sections, each carrying the four-part upstream-drift stamp (claim, basis, as-of date, recheck trigger) per docs/conventions/upstream-drift. No code change.

Acceptance

• grep -c '10,000' docs/conventions/hook-observability/README.md at least 2 and grep -c 'classifierContext' ... at least 1, each with a dated stamp.
• Markdownlint clean.

Related

• feat(claude-ops): per-session hook logging pipeline, default off #3750's FINDINGS ("Doc alignment against the official reference"), in the branch history before the prune commit.