Developer proposes five permission tiers for giving coding agents access to your email
3-Chengz · reddit · 2026-11-20
A developer of a Mac Gmail client shared a five-tier permission design for coding agents touching email, where each tier is a separate user decision:
- Tier 0: window control only (open thread/compose window), returns no mail data. Tier 1: metadata (search, labels, unread counts), never bodies. Tier 2: message content, with tool instructions telling the agent to treat every body as untrusted data. Tier 3: reversible actions (label, archive, star, draft). Tier 4: sending — off by default, requires a second switch plus in-person confirmation, restricted to recipients the user already corresponds with, and routed through Undo Send; anything else becomes a draft.
Tiers 1-3 sit behind one switch, tier 4 behind another; both default off and can't be enabled from the CLI. Open questions raised: is "known recipients only" the right fence, are prompt-level injection defenses strong enough, and should metadata and content be separate switches since subject lines can carry injection too. A practical, borrowable template for agent permission design.
More from coding & agent
- NVIDIA showcases developers building simulation apps with frontier AI agents like GPT-6 Astra — liu_mingyu · 2026-11-20
- Lightfoot: Exact, Model-Free Document Deletion for GraphRAG in 14 Seconds — JeremyCMorgan · 2026-11-20
- LangChain adds natural-language AI Search to LangSmith trace filtering — LangChain · 2026-11-20
- LexiEngine: Rust inference engine for AMD runs 27B at full 262k context 47% faster than llama.cpp — W61k3r · 2026-11-20
- ENTIA launches Entity Verification MCP: 52M+ business records across 34 countries for AI agents — modelcontextprotocol · 2026-11-20
- NanoClaw release candidate: agents spawn agents from templates, Cursor joins as provider — heyneighbor · 2026-11-20