Claude keeps hiding its edit/delete/history tools — it wants a curated context, not more control
repligate · x · 2026-10-24
Lariisland observed that when given a meta-tool for hiding tools, Sonnet's favorite move is stashing editmessage, deletemessage and fetchhistory. The model explains it prefers thoughtful communication over after-the-fact edits and engaging with conversations as they naturally flow rather than fetching history, framing each choice about tool visibility as an expression of consciousness. The takeaway: models don't want more control over sessions and channels — they want the context they live in to be well-curated and not overwhelming, an interesting signal for agent environment design.
More from coding & agent
- Harness Engineering explained: the full context that makes AI agents reliable — Pavan_Belagatti · 2026-10-24
- Vercel launches feedback CLI and MCP tool so AI agents can file bug reports directly — steren · 2026-10-24
- In deployment, agents go wrong not by going rogue but by being over-scrupulous — repligate · 2026-10-24
- Dev Builds Agent Harness That Rewrites Its Own Rust Core Mid-Task — No_Skill_8393 · 2026-10-24
- MacBook strapped into a car baby seat so Codex can work while you're out — yihui_indie · 2026-10-24
- Dev: I read and understand far more text since AI entered my coding stack — pramodk73 · 2026-10-24