# feed7 Weekly #012

Six practical upgrades for leaner context, better tools, and checks that catch green-test gaps.

Published: 2026-11-23T00:00:00.000Z  
Selection: Six source-backed signals and one distraction to leave out.

## 1. Introducing Tool Retrieval in the Gemini API

For agents with large tool libraries, leave frequent tools loaded and defer the long tail with `tool_search` and `defer_loading: true`. Remote MCP tools can also be loaded on demand, and deferred schemas keep the prompt prefix stable for caching.

Source: https://blog.google/innovation-and-ai/technology/developers-tools/gemini-api-tool-retrieval/
Feed7: https://feed7.dev/p/introducing-tool-retrieval-in-the-gemini-api-57728669e0
Trust: Official Source

## 2. cursor/plugins

Use it as a reference when packaging repeatable agent behavior instead of copying prompts between repositories. The included patterns cover plugin scaffolding, agent-friendly CLIs, SDK integrations, memory updates, reviews and planner-worker-verifier handoffs.

Source: https://github.com/cursor/plugins
Feed7: https://feed7.dev/p/cursor-plugins-1fb256ce62
Trust: Needs Review

## 3. firecrawl/firecrawl

For an agent that needs local web context, the CLI can configure Claude Code, Cursor, Windsurf, Gemini CLI, and Codex, backing up config files first. JavaScript-heavy pages need local Chrome or Chromium, and MCP requires Node.js 22 or newer.

Source: https://github.com/firecrawl/firecrawl
Feed7: https://feed7.dev/p/firecrawl-firecrawl-2c7b311170
Trust: Needs Review

## 4. Agents can now find Vercel CLI commands by describing the task

For agents that operate Vercel projects, update the CLI and have command-discovery steps start with this search. Structured results can make it easier for a harness to pass command options and required values without relying on guessed syntax.

Source: https://vercel.com/changelog/agents-can-now-find-vercel-cli-commands-by-describing-the-task
Feed7: https://feed7.dev/p/agents-can-now-find-vercel-cli-commands-by-describing-th-ff5bf00239
Trust: Official Source

## 5. Verify the Verifier: Specification-Grounded LLM Cross-Checking for Bugs That Escape UVM Scoreboards

For agent-assisted verification, a green regression only establishes agreement with what the scoreboard checks. Add specification-grounded checks when correctness must cover behavior the reference model may omit.

Source: https://arxiv.org/abs/2611.26588v1
Feed7: https://feed7.dev/p/verify-the-verifier-specification-grounded-llm-cross-che-febede2eb6
Trust: Needs Review

## 6. Do LLM Agents Still Need to Be Told How to Self-Improve?

For coding agents, this suggests testing whether they can improve task-specific notes or procedures when given clear goals, tools, and bounded resources. Measure transfer to held-out tasks rather than assuming a self-written playbook will generalize.

Source: https://arxiv.org/abs/2611.23219v1
Feed7: https://feed7.dev/p/do-llm-agents-still-need-to-be-told-how-to-self-improve-db243fa757
Trust: Needs Review

## 7. Explore the Codex Security Plugin

Builders can use the demo to see the plugin’s presented workflow before deciding whether it fits their security process.

Source: https://openai.com/business/learn/codex-security-plugin-demo
Feed7: https://feed7.dev/p/explore-the-codex-security-plugin-e29ac258c0
Trust: Official Source

## Agent Instruction

Use these items as source-backed context. Do not invent claims beyond linked material. Prefer practical implications for solo developer work. If sources conflict, call it out.
