Fireworks prompt caching is enabled by default (automatic prefix matching),
but on serverless infrastructure, requests hit random replicas. Without
session affinity, the per-replica cache misses, negating cache hit rates
and the discounted cacheRead pricing.
Changes:
- Add sendSessionAffinityHeaders and supportsCacheControlOnTools
to AnthropicMessagesCompat interface
- Send x-session-affinity header for Fireworks (and Cloudflare AI
Gateway Anthropic) when sessionId is available and caching is enabled
- Omit cache_control on tool definitions for Fireworks (unsupported
per https://docs.fireworks.ai/tools-sdks/anthropic-compatibility)
- Default supportsEagerToolInputStreaming to false for Fireworks
(unsupported field)
- Default supportsLongCacheRetention to false for Fireworks
(cache_control.ttl not supported)
- Add compat settings to Fireworks models in generate-models.ts
- Update generated models with Fireworks compat settings
- Add integration tests for session affinity and tool compat
Refs: https://docs.fireworks.ai/guides/prompt-caching
Refs: https://docs.fireworks.ai/tools-sdks/anthropic-compatibility
Skill slash commands store a structural <skill>...</skill> wrapper in raw
user messages. The TUI uses parseSkillBlock() to split this into separate
SkillInvocationMessageComponent and UserMessageComponent siblings, but the
HTML export renderer passed the full raw text through markdown, causing
broken/dangling XML tags to appear in exported HTML.
Add parseSkillBlock() to the export template and render skill-invocation
and user-message as separate sibling blocks:
- Sidebar tree shows skill name + user prompt separately
- Content area shows a clickable skill-invocation block (collapsed by
default, markdown content on expand) followed by the user message
- Copy-link button preserved on the wrapper element
- Toggle tools (O key) expands/collapses skill invocations alongside
compaction and tool output blocks
