refactor(webex): converge on the shared nonce minter and byte chunker

[bolichen97]

Three duplications of a shared messaging helper, all in code #5218 just landed. Found by auditing that branch's own new code against the messaging layer rather than by a review finding.

The options card minted its own nonce

webex/renderer.py used a local secrets.token_hex(8), while every other channel with a clickable control — Discord, Teams (×2), Telegram — and both shared registries (PendingApprovals.reserve, PickerRegistry.mint) reach messaging.renderer.new_approval_nonce().

Being honest about severity: entropy is identical (both 8 bytes / 64 bits), so no token was weaker and this is not a live security hole. What it violates is that helper's stated contract:

Every channel with a clickable approval has to mint one, compare it on resolve, and retire it with the prompt — and three independent copies of that is how one of them ends up with a weaker token or none at all.

The alphabet and length do differ (16 hex chars vs ~11 urlsafe), and the shared 8-byte choice exists because it has to fit Telegram's 64-byte callback_data — a constraint the next person to touch a local copy has no way to know.

Worst of it: the file disagreed with itself. The approval card on this same renderer already routes through the shared minter via PendingApprovals.reserve, with a comment saying so; the options card 380 lines later rolled its own.

outbound_mimetype was dead on arrival

No caller anywhere in src/ — only its own three tests, which gave it false life (coverage, plus "it's tested, so something must use it").

It was also dead for a good reason: it derived the content type from the path extension, which is what the outbound-files contract forbids —

Only a real raster is sent. Type comes from the leading bytes via messaging/raster.py, never an extension.

— and the upload path correctly uses item.mime, the sniffed type. Removed, with its tests and two now-unused imports.

chunk_utf8 re-implemented the shared primitive, minus its guards

webex/client.py::chunk_utf8 duplicated messaging.split.chunk_utf8_bytes loop-for-loop, and was missing both of the shared version's termination guards:

	shared chunk_utf8_bytes	the Webex copy
max_bytes <= 0	return [text]	spins forever
a code point wider than the budget	falls back to remaining[0]	spins forever

So the copy was the strictly worse one, and its failure mode is a hang rather than an exception. Not reachable today — the single caller passes the 7000-byte default — so this is a latent hazard, not a live bug.

The local wrapper is now gone entirely and renderer.py calls chunk_utf8_bytes(content, WEBEX_MAX_TEXT) directly. An earlier revision kept it as a one-line delegating wrapper on a claimed symmetry with truncate_utf8 above it; that symmetry does not hold and the claim is withdrawn — truncate_utf8 earns its wrapper with three same-file consumers (client.py:447, :494, :522), while this one had zero, its single consumer lives in renderer.py and always took the default, and test_webex_client.py was already importing and exercising the shared primitive, so the local copy had no test of its own. The guard and fence-blindness rationale moved to the one call site where the choice is actually made.

Semantics preserved deliberately: still the fence-blind primitive, not split_markdown_bytes. Webex's answer path re-seals its own fences and is pinned that way by test_channel_table_rendering.TestDeliveryFraming.

Riding along, declared

test_webex_attachments.py is not purely a deletion. Removing outbound_mimetype took the file's only tests with it, so the same commit adds TestProcessWebexAttachments (3 tests) plus a bodies= download capability on the file's FakeClient — backfilling the only coverage of the ingest wiring at transport_dispatch.py:458. Test-only and coverage-positive, but an addition inside a subtraction PR, so it is stated here rather than left to be discovered in the diff. Its two temp-producing tests keep cleanup(result.temp_paths) in a finally so a failed assertion cannot leave downloaded bytes behind.

Not in scope

The [OPTIONS:] trailer parse is duplicated across seven sites (Discord, Telegram, Teams, WeCom, Slack, Webex, and render_options_as_text itself). That is a real convergence but it touches six channels, so it goes in its own PR rather than riding along here.

Verified locally: 504 webex-selected tests pass, plus the framing/split pins (test_channel_table_rendering.py, test_webex_client.py, test_messaging_split.py, test_webex_attachments.py — 316 passed); mypy (1104 files), flake8, isort and the black baseline ratchet all clean. Rebased onto current main; one commit.

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of 12bdca3c48ac30ee652502a076d5c100238a637b — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Three verified convergences onto owned primitives; behavior preserved in every reachable case (empty-input [], opaque nonce round-trip), and the ride-along tests are declared.

[DESIGN-REVIEWED] 12bdca3

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed 12bdca3c48ac30ee652502a076d5c100238a637b — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 12bdca3

Verdict parsed from the review's SHA-scoped output markers for commit 12bdca3c48ac30ee652502a076d5c100238a637b.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 12bdca3c48ac30ee652502a076d5c100238a637b: <one-sentence reason>

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 12bdca3c48ac30ee652502a076d5c100238a637b and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 12bdca3

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 12bdca3c48ac30ee652502a076d5c100238a637b: <one-sentence reason>

[github-actions]

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of 12bdca3c48ac30ee652502a076d5c100238a637b — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All claims verified. This is a pure-subtraction PR: three duplications removed in favor of existing shared mechanisms, with one declared test-only rider. Final review:

First-Principles-Verdict: PASS

Every item deletes a second spelling of a mechanism that already exists in messaging/, and the one rider is declared, test-only, and replaces coverage the deletion removed.

What this change ships

Intent: make Webex use the shared messaging helpers instead of three private copies — a FIX (of duplication), not an addition.

1. Options card nonce now comes from the shared minter — justified (6 other call sites already use new_approval_nonce; nonce is opaque, compared by compare_digest at cards.py:176, so the alphabet change has zero format consumers)
2. outbound_mimetype deleted with its tests — justified (grepped repo-wide: 0 consumers remain; only its own tests ever called it, and it derived type from extension against the sniff-only contract)
3. Local chunk_utf8 deleted; renderer calls shared chunk_utf8_bytes directly — justified (shared version at split.py:350 carries both termination guards the copy lacked; 0 remaining references in src/ or test/)
4. TestProcessWebexAttachments (3 tests) + FakeClient.bodies= — rides along, declared in the description; test-only, backfills the ingest coverage the deletion took
5. truncate_utf8 docstring rewritten to name the shared splitter — rides along, trivial

Watch

The description defers the [OPTIONS:] trailer convergence; I count 6 renderer call sites of split_options_trailer plus the shared default — the deferral is real and correctly sized out of this PR, not an omission.

[FIRST-PRINCIPLES-REVIEWED] 12bdca3