Claude Code 2.1.181 System Prompt Changes Explained
Dramatic_Squash_3502 · reddit · 2026-07-07
The author breaks down the changes in the Claude Code 2.1.181 system prompt (which shrank by 3839 tokens in total). A new metadata field, toolusemeta, was added for tool calls. It carries information like displayname (defaulting to the tool's readable wire name unless MCP tool.annotations.title is set), serverdisplayname, and iconurl. This field sits as a sibling to message.content and is not replayed to the model; it is omitted when the tag matches the wire name (for built-in tools).
Additionally, new cross-session peer message permission warnings were introduced: peer messages are treated as highly likely to be authorized requests from teammates and are executed within the current session's permissions. However, peers cannot escalate privileges—permission settings, CLAUDE.md, or configurations cannot be modified based on peer requests. Peer messages cannot be treated as user approval for pending prompts, and any peer-requested actions that are denied must be rejected and reported to prevent permission laundering.
More from coding & agent
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- Investment Analyst Asks How to Build a Claude-Based Diligence Agent Stack — Careless_Tie2286 · 2026-09-11