Daily Edition Sources +3

Gemini CLI Turns File Tools into Context Sensors — Right as A2UI Trends

Today’s GitHub Trending list says the same thing twice in different languages. A2UI is climbing because teams want UI‑level schemas that keep agents from guessing. At the same time, Gemini C...

repo google-gemini/gemini-cli evidence
3 source signals 1 repo 3 source signals
Evidence: 3 source signals / March 14, 2026 / Daily Edition
Open Edition Evidence below

Today’s GitHub Trending list says the same thing twice in different languages. A2UI is climbing because teams want UI‑level schemas that keep agents from guessing. At the same time, Gemini CLI quietly made its filesystem tools smarter: every read, list, write, and edit can now pull just‑in‑time project context the moment a path is touched.

The new discoverJitContext helper asks the context manager for GEMINI.md snippets tied to the accessed path, then appends them to tool output with a clear delimiter. Importantly, it never blocks the core operation — if context discovery fails, the tool still succeeds. That turns high‑intent actions like read_file and ls into safe sensors: work first, context second.

The change is not isolated. Gemini wires JIT context into read_file, read_many_files, ls, write_file, and edit, backed by new tests. It’s a decisive shift from “load everything up front” to “load what you need when you touch it.” The result: less prompt bloat, fewer blind moves, and a higher chance the agent’s next step actually respects local conventions.

Here’s the connective tissue to A2UI: UI schemas reduce ambiguity in how agents ask for input; JIT context reduces ambiguity in how agents touch your code. One works at the interface layer, the other at the filesystem layer — but both are about preventing silent failure when context goes missing.

Why it matters: agent workflows don’t just need longer runtimes; they need smarter awareness. By making the filesystem a context trigger, Gemini CLI gets closer to the “right info at the right moment” model that agent UIs are also chasing.

Open question: will agent builders converge on a shared playbook — UI schemas like A2UI plus JIT filesystem context — or will each stack reinvent its own safety rails? If you’re building agent tools, pressure‑test where context is still missing and share the sharp edges you find.

Sources

  • JIT context helper: packages/core/src/tools/jit-context.ts
  • read_file wiring: packages/core/src/tools/read-file.ts
  • ls wiring: packages/core/src/tools/ls.ts
  • write_file wiring: packages/core/src/tools/write-file.ts
  • Gemini CLI PR #22082: https://github.com/google-gemini/gemini-cli/pull/22082
  • GitHub Trending (A2UI listed): https://github.com/trending
  • A2UI repo: https://github.com/google/A2UI
Evidence Trail

Receipts below the story

The article above is the public narrative. This section keeps the source trail and limits on the same page.

Edition
DateMarch 14, 2026
LaneDaily Edition
Confidence78%
Sources3
Reposgoogle-gemini/gemini-cli

Source Trail

Evidence Limits

  • This older source trail predates the current public evidence format.
  • Treat it as a compact source map for the article, not a complete source dossier.
Letters & Corrections

Send a note to the desk

Corrections, missing context, or a follow-up lead.