public/galaxy
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Filter unsupported reasoning from Bedrock request history

Browse Source
This commit is contained in:
Ryan Ward
2026-10-28 18:26:02 +00:00
parent 3cce041a6d
commit f399ada2a3
7 changed files with 462 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
AGENTS.md
+1
View File
@@ -223,6 +223,7 @@ Key invariants:
- `recall_tool_history` excludes earlier calls to itself; archived tool results remain searchable by query or exact `tool_use_id`
- Before progressive summarization drains messages, `ConversationMessage::archive_tool_results()` moves tool-use/result pairs into `tool_result_archive`
- Bedrock prompt caching uses three cache points: system prompt, second-to-last history message, and tool configuration
- Bedrock requests project reasoning history for the selected foundation model: omit unsigned response-only reasoning and signed blocks from incompatible models while preserving Claude's complete signed blocks for tool continuations. Keep durable/UI history intact, remove messages emptied by projection, and use the discovered foundation model when invocation targets an opaque inference-profile ARN.
- `ensure_tool_results_paired()` enforces Bedrock's invariant that every `tool_use` has a matching `tool_result`
- Progressive summaries are prepended to provider requests as a user/assistant pair; direct-provider runs compact their live transcript at a checkpointed `ReadyToCallModel` boundary before another model call, while background summarization remains independent for session-owned runtimes
- Direct-provider progressive compaction runs in a separate request alongside foreground work. Start early against the smaller usable history budget of the current and terminal-use profiles, including each profile's prompt/tools/output reservation. Apply a result only at a safe model boundary after validating its exact snapshotted prefix, preserving all newer history and recomputing live context accounting. Wait for pending compaction only after context exhaustion; cancellation and terminal cleanup must remain immediate. Cleanly completed conversations may accept a late summary while idle; stale, cancelled, or superseded generations must never replace a newer run's history.