feat(web): share conversation attribution across agent applications

[drewstone]

Shared attribution in the existing browser-safe foundation

Extract SUPER's generally useful assistant-message grouping into /web. One pure helper, not another transcript store, model loop, framework or execution state machine.

groupConversationMessages groups successive assistant messages until a real user message, speaker change or conversation change. Tool/progress notices do not repeat attribution. Streaming/working rows can share a group. Original bodies, IDs, timestamps and objects remain intact; accessible attribution stays the renderer's responsibility.

The existing web implementation is moved byte-for-byte into core.ts behind its original entry point; the barrel exports all previous helpers plus the attribution helper. No new dependency, peer-floor change, public package subpath or manual version bump.

Documentation

docs/conversation-attribution.md covers usage, speaker/thread boundaries, pagination, streaming and accessibility. CHANGELOG.md records Unreleased changes. The generated API documentation/codemap has been regenerated and checked, not left stale. The temporary branch-scoped regeneration workflow was removed after it completed; there is no added write-capable workflow in the final branch diff.

Full-checkout verification passed

At cbaffe49cf1cee16b54b062c96ec4a168c696217:

• Frozen install and release-workflow contract.
• Generated documentation freshness and full typecheck.
• 391 gate tests passed, including browser-safe exports, streaming, sandbox boundaries and test quality.
• Full build.
• 6,224 source tests passed, zero failed; 10 pending/skipped tests remain in the repository suite. The attribution tests use native Vitest assertions and all nine pass.
• Generated application checks, including installed peer validation, typecheck, 15 chat-template tests and a Wrangler dry run.
• Knip check passed (informational hints remain).

Evidence: https://github.com/tangle-network/agent-app/actions/runs/35031975226 (artifact pr-verification-594 contains source, exact commit and logs/results).

Full-checkout checks initially caught stale generated docs and strict test-indexing/assertion conventions. Those were corrected on this same PR; the checks were not disabled or replaced with offline shims.

Actual consumer cutover

https://github.com/drewstone/super-agent/pull/4 replaces its independently maintained grouping body with a hash-verified generated build of this canonical source. Both its existing renderers use that build. Until a published compatible package exists, exact source/build pins keep the app runnable without an invented release version or a second maintained fallback.

Related provider primitives: tangle-network/agent-integrations#304.

Non-draft and ready for review. This does not claim a React migration in SUPER, actual browser/device usability qualification, a live model benchmark, npm publication, or a production deployment. No default AgentProfile or permission policy is changed.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.