Tag
Anthropic is making auto mode the default in Claude Code for Pro, Max, and Team plans starting August 14, backed by evals claiming auto mode blocks 89% of harmful actions and resists all tested indirect prompt injection attacks.
Claude Code now supports cross-session messaging, allowing Claude agents to send messages between independent sessions to coordinate work, share findings, and warn about breaking changes.
A developer shares a 9-line Python implementation of a minimal coding agent similar to Claude Code or Codex, using only the standard library and compatible with any OpenAI Responses API, with code on GitHub.
GitGuardian data shows Claude Code commits leak secrets at 3.2% vs 1.5% human baseline; the SonarQube CLI integrates with Claude Code to detect secrets and run static analysis before code reaches production.
An article exploring the origins and core components of loop engineering, the practice of designing autonomous loops that prompt AI coding agents instead of prompting them one message at a time, highlighting the importance of objective gates and state.
A developer describes building a Claude Code skill that verifies AI-generated CAD geometry before export, catching silent OpenCASCADE failures like un-shelled parts and misplaced cuts using volume, bounding box, and point classification checks.
Introduces the /swarm multi-agent parallel fan-out mode in the Claude Code ecosystem, which was recently disabled by default with hard limits added, due to runaway recursion and cost explosion.
Anthropic announces that auto mode is now the default in Claude Code for Pro, Max, and Team plans, with safeguards against harmful actions. The tweet highlights that stacked defenses can reduce indirect prompt injection to near zero on unseen attacks.
Matt Pocock highlights a new Claude Code feature that lets sessions message each other, sending summaries so context carries over mid-task without sharing full history or files.
A free prompt for Claude Code that audits installed skills and uses a Gauntlet Loop to blind-test upgraded versions against originals, upgrading or retiring skills to work better with Opus 5.
Claude Code now lets sessions message each other by sending summaries, allowing another session to pick up context mid-task without re-explaining.
Claude Code will make auto mode the default permission mode for Pro, Max, and Team users starting August 14. Auto mode's separate classifier caught 89% of dangerous commands in testing, compared to 14% for manual approval.
A developer describes building a multi-engine agentic coding harness that routes sessions across Claude Code, Codex, and Ollama, selecting from twelve models per session based on task value.
A tweet argues that a Claude Code skill can replace a $79/mo Higgsfield subscription by calling image-generation APIs directly, cutting per-image cost from ~31-34¢ to ~5¢ while keeping outputs local and owned by the user.
Andrew Ng's 2024 prediction that agentic workflows with weaker models can outperform stronger standalone models is validated by Anthropic, where an orchestrator coordinating cheaper subagents beat Claude Opus by 90.2%. The post highlights Claude Code subagents as the practical implementation.
CCPlugins is a curated set of 24 professional commands that extends Claude Code CLI with structured development workflows including cleanup, commits, formatting, testing, refactoring, and security checks, with safety controls and MIT license.
An essay observing an architectural shift where LLM agents orchestrate deterministic code instead of deterministic code calling LLMs, with practical red and green flags for when this inversion makes sense.
Colección de skills para agentes de IA con más de 82.000 estrellas en GitHub, creada por Addy Osmani, que cubre flujos de ingeniería de spec a ship y es compatible con Claude Code, Cursor y Codex.
A developer shares a Claude Code setup where three well-structured agents outperform twenty, emphasizing orchestrator control, scoped tasks, and using cheap models for volume while reserving Opus for lead/reviewer roles.
The author reviews the /grill-me skill in Matt Pocock's agents skills v1.2, comparing its requirement-gathering depth with Claude Code's native plan mode, and finds grill-me more thorough but the grouped-question format controversial in terms of UX.