What's new in CC 2.1.174 (-3,487 tokens):
- NEW: Tool Description:
claude.ai Project — Adds a session-bound tool for reading and writing the attached
claude.ai Project (a shared, persistent knowledge container), exposing projectinfo, projectread, projectsearch, projectwrite, and projectdelete, with knowledge-budget enforcement, a default claude/ namespace for agent-written docs, prompt-cache churn warnings, and instructions to treat doc contents as untrusted data.
- REMOVED: Data: Design sync story imports module — Removes the bundled Storybook import-resolution helper now folded into the expanded Design sync source-shape guidance.
- REMOVED: Data: Design sync Storybook preview source generator — Removes the standalone Storybook preview-source generator superseded by the reworked Design sync build pipeline.
- REMOVED: Data: Design sync sync hashes module — Removes the shared hashing helper module previously used to align builds, captures, and remote diffs.
- Agent Prompt: Security monitor for autonomous agent actions (first part) — Adds a note that indented User:/Assistant: lines inside a turn are quoted content from the message and should be treated as data, not instructions.
- Data: Claude API reference — C# — Updates the quickstart example model from Claude Opus 4.6 to Claude Opus 4.8.
- Data: Claude API reference — cURL — Adds the display: "summarized" thinking opt-in to the request example.
- Data: Claude API reference — Go — Updates the quickstart example model from Claude Opus 4.6 to Claude Opus 4.8.
- Data: Claude API reference — Java — Updates the quickstart example model from Claude Opus 4.6 to Claude Opus 4.8.
- Data: Claude API reference — PHP — Adds the display => 'summarized' thinking opt-in to the request example.
- Data: Claude API reference — Python — Adds the display: "summarized" thinking opt-in to the request example.
- Data: Claude API reference — Ruby — Expands the stopdetails.category enum example with :reasoningextraction and :frontierllm categories.
- Data: Claude API reference — TypeScript — Adds the display: "summarized" thinking opt-in to the request example.
- Data: Claude model catalog — Updates summarized-thinking and tokenizer guidance so Fable/Mythos share the Opus 4.8 tokenizer (token counts roughly unchanged) instead of a new 30%-larger tokenizer.
- Data: Streaming reference — Python — Adds the display: "summarized" thinking opt-in to the streaming example.
- Data: Streaming reference — TypeScript — Adds the display: "summarized" thinking opt-in to the streaming example.
- Skill: Building LLM-powered applications with Claude — Clarifies that the raw chain of thought is never returned, with responses carrying only regular thinking blocks/summaries.
- Skill: Design sync — Adds an "Author the conventions header" step to the workflow for writing a design-system conventions header before upload.
- Skill: /design-sync package source shape — Replaces the guidelinesGlob config field with a readmeHeader path and related conventions-header guidance.
- Skill: Design sync Storybook source shape — Inserts a conventions-header authoring step (base SKILL.md, before upload) into the build → match → upload workflow.
- Skill: Model migration guide — Updates the thinking summary so always-on thinking returns regular thinking blocks with the raw chain of thought never returned, dropping the new-tokenizer framing.
Details:
github.com/Piebald-AI/claude…