Skip to content

Mod/mod 3 anthropic adapter - #20

Merged
SaketMunda merged 6 commits into
masterfrom
mod/MOD-3-anthropic-adapter
Oct 10, 2026
Merged

SaketMunda merged 6 commits into
masterfrom
mod/MOD-3-anthropic-adapter

Conversation

@SaketMunda

Copy link
Copy Markdown
Owner

Module

Brief: roadmap/modules/<ID>-*.md

What changed

Definition of Done

  • Every item in the brief's Acceptance criteria is met and demonstrated (test or recorded manual check).
  • scripts/verify.sh is green (typecheck, lint, unit tests).
  • New logic in packages/* has unit tests (target ≥80% line coverage on new files). UI logic (state, reducers, parsers) is tested. Pure presentation isn't required to be.
  • No any / as unknown as at package boundaries. Every external input is validated with zod.
  • No new network destinations except ones the user configured. No telemetry.
  • Errors are handled: user-facing failures show an actionable message, and nothing fails silently.
  • Brief updated: Status, Handoff notes (what was built, deviations, known gaps, follow-ups).
  • PR opened from mod/<ID>-<slug>, description links the brief, CI green.

Acceptance criteria evidence

# Criterion Evidence

SaketMunda and others added 6 commits October 10, 2026 18:06
…event

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…yFor

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ollama: message.thinking -> reasoning_delta, reasoning -> think (only for
models with the thinking capability). OpenAI-compatible: reasoning_content /
reasoning deltas, reasoning -> reasoning_effort, dropped if the server
rejects it. Unset sends nothing.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Message mapping (alternating turns, tool_result pairing), streaming with
tool use and merged usage incl. cache tokens, cache breakpoints on the system
prompt and cacheable messages only, model table for capabilities and cost,
error mapping (overloaded -> rate_limit, prompt too long -> context_length),
extended thinking per reasoning level with signed blocks round-tripped via
providerState. The SDK never resolves ambient credentials or logs to stdout.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
reasoning_delta becomes a task event with the turn's messageId; provider
state rides on the assistant turn and historyFor drops it for another model.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@SaketMunda
SaketMunda merged commit a916617 into master Oct 10, 2026
1 check passed
@SaketMunda
SaketMunda deleted the mod/MOD-3-anthropic-adapter branch October 10, 2026 13:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant