Steven Gonsalvez

Software Engineer


CEREBRO


machine-read, human-curated

Researched and written with 265,742 tokens (in 8 · out 37,240 · cache-read 50,794 · cache-create 177,700) across 4 Claude calls

Coding agents: Claude Code and context economics

  • mksglu/context-mode sandboxes MCP tool output (claims 98% reduction, e.g. a 56 KB Playwright snapshot) and persists session memory across compaction, via MCP plus hooks on 17 platforms.
  • v2.1.299 adds idle compaction (200K+ token conversation idle ~54 min is compacted before the 1h prompt cache expires) and CLAUDE_CODE_DISABLE_FILE_MENTIONS (@ mentions pass paths, not contents).
  • v2.1.300 adds cache_creation_5m_tokens and cache_creation_1h_tokens to the OTel api_request event, so you can see cache writes split by TTL; also adds mod virtualText for prompt decorations.
  • v2.1.301 adds CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS so unattended retry mode gives up on usage limits that reset too far out, plus OTel tracing guidance in the claude-api skill.
  • xerj-org/xerj is a local index over code, docs, logs and PDFs with an Elasticsearch-compatible API, claiming 40x fewer tokens than grep for agent retrieval.

Agent orchestration and infrastructure

Agent harnesses, plugins and frameworks

  • obra/superpowers is a skills-based methodology that makes the agent clarify requirements before it writes code.
  • affaan-m/ECC is a harness optimisation layer (skills, memory, security) spanning Claude Code, Codex, Opencode and Cursor.
  • ruvnet/ruflo is a swarm and orchestration harness installable as Claude Code plugins (core, console, mods); needs Claude Code 2.1.287+.
  • anthropics/claude-plugins-official is Anthropic's curated plugin directory, with a warning that it doesn't vet what plugins ship.
  • b-nnett/codex-plusplus is a macOS tweak and extension system for the Codex desktop app, patching the main process outside the app bundle.

Agent clients and CLIs

Agent-driven media and SaaS

Signals

  1. mksglu/context-mode: Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and…
  2. v2.1.300
  3. v2.1.301
  4. v2.1.299
  5. obra/superpowers: An agentic skills framework & software development methodology that works.
  6. Harness by Autonomous
  7. How Anthropic’s Dynamic Workflows Coordinate Claude Agents
  8. anthropics/claude-plugins-official: Official, Anthropic-managed directory of high quality Claude Code Plugins.
  9. RT @ClaudeDevs: We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side. It's now muc…
  10. Found a GitHub repo that does Claude Code’s job without for free. Yesterday Accenture told staff it will only pay for $29 a month of Claude…
  11. affaan-m/ECC: The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Cl…
  12. 0.163.0
  13. simonw/llm: Access large language models from the command-line
  14. Redesigning source control at Anthropic
  15. b-nnett/codex-plusplus: Codex++ tweak system for the Codex desktop app
  16. RT @AnthropicAI: New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In Sep…
  17. ruvnet/ruflo: 🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversatio…
  18. Quoting Geoffrey Huntley
  19. Rushilo
  20. Code users in the Claude Desktop app rely on folder groups, visible context, safe permission defaults and more (summary based on GitHub iss…
  21. xerj-org/xerj: XERJ is the new way for AI to search data. Its autoindex capability activates agents to know your data without the token was…
  22. I built a full DaVinci Resolve-style video editor where Claude Code is the motion designer — open source
  23. heygen-com/hyperframes: Write HTML. Render video. Built for agents.
  24. RT @jullerino: Orchestrator V2 is now out on the stable channel in @t3dotcodes! We also thought it was time we drop the Alpha branding, so…
  25. RT @theo: @thsottiaux - Skills that route to different behaviors based on model - Skills that have recs like "don't use slow git command" w…

Sources & citations

rss1409
github349
x2225
reddit252
show_hn400
hackernews300
github_search200
swipe200
yc-rfs130
gmail80
yc-launch20
seed_urls10
crackscan00
ossinsight00
reddit_users00
total55525