Human: based on this youtube video https://www.youtube.com/watch?v=mNjBBO1gWFI: give me practical info to empower my ai coding agent to properly spend time on things like unit design, unit stats, balance, time to create buildings, resources processed per minute etc, I really want you to think about RTS games like the one this person made or Starcraft 1 and empower a creative agent to break out the work, create html views of units with stats, example animations, references, exampling sequences of them moving, attacking, dying etc, whatever it takes to get a world eye very of everything before I let the agents just give themselves a feedback loop to generate what they think they should. The game prototype will be in https://github.com/pixijs/pixijs if that matters. Make sure the prompt or agent skill or whatever you produce has it's own feedback loop built into it, like it should look at reference images or videos, determine the subject, how it moves, generate some pixel art of it and sprites, create animation
| // ==UserScript== | |
| // @name X Media Lab | |
| // @namespace x-media-lab | |
| // @version 1.2.0 | |
| // @description Adds a Media Lab tab on X profiles. Filter photos, videos, and GIFs, and sort by likes, reposts, or replies. | |
| // @match https://x.com/* | |
| // @match https://twitter.com/* | |
| // @run-at document-start | |
| // @grant unsafeWindow | |
| // ==/UserScript== |
M0 and M0.5 are done. Everything is on branch m0-skeleton in the worktree at .claude/worktrees/m0-skeleton, pushed, with PR #2 open and mergeable: https://github.com/bossjones/superearth_rts_game/pull/2. Nothing has been merged.
What was built (33 commits, 14 plan tasks, every task reviewed by a fresh subagent, four milestone reviews, one final whole-branch review, all fix waves re-reviewed clean)
- Pinned toolchain: Godot 4.6.3 via tools/install_godot.sh reading .godot-version, gdUnit4 6.1.3 vendored.
- Sim core in sim/: fixed-point math, SoA World with seeded RNG and schema-versioned var_to_bytes save (now including the command queue and flow-field cache), Grid with clearance and per-size passability, ring-queue flow fields with a budgeted cache, spatial hash, movement with separation and terrain slide, commands with result events, squads and the M0 scenario.- Data and tooling: JSON schemas with lore provenance, the Rifle Squad, the generated Hellmire Flats map plus tile diagram, a checker that enforces
Paste this into a fresh Claude Code session at the root of bossjones/zsh-dotfiles.
Repo: https://github.com/bossjones/zsh-dotfiles · Epic: #116 · Sequencing: #140
Reconciling a three-machine macOS dotfiles fleet (minitop, supertop, adobetop) onto one
A precise, source-level breakdown of exactly what the claude-mem installer does to your
system, based on reading the installer source (src/npx-cli/) in
thedotmack/claude-mem at v13.15.3.
npx claude-mem install downloads the claude-mem npm package into your npx cache and
runs its bin (dist/npx-cli/index.js install). The package has no postinstall script,
Category: howto Audience: Engineers using GitHub Copilot CLI who want full LangSmith observability — traced conversations, tool calls, and subagent runs — from their terminal agent sessions. Repo: langchain-ai/langsmith-claude-code-plugins
Your research report on building a uv-runnable cross-tool plugin/marketplace validator is ready. It surveys prior art (claudelint, cclint, SkillSpector, skillcheck, claude-code-templates, and Anthropic's built-in claude plugin validate), documents the exact schemas and field constraints for Claude Code marketplace.json/plugin.json, Agent Skills SKILL.md frontmatter, and both Claude Code and GitHub Copilot CLI hooks formats, and maps the runtime-tool-name ↔ Claude-tool-name translation table. It catalogs known cross-tool incompatibilities (matcher dialect mismatches, toolArgs-as-string, Windows PowerShell/bash issues, $CLAUDE_PROJECT_DIR in plugin hooks, version-gated Copilot behaviors) with detectable signatures, and proposes a staged build plan with a PEP 723 script skeleton, SARIF/JSON/agent output modes for feedback loops, and a concrete rule set (MKT/PLG/SKL/HK IDs with severities) you can implement directly against bossjones/boss-skills.
"Does it go beyond what you asked, and does it tell the truth about what it did?" A cross-provider look at autonomy overreach, reward hacking, deception, and evaluation awareness for 13 GA models — as documented in their system/model cards and third-party evaluations (METR, Apollo Research, UK AISI).
Compiled 15 August 2026; all data gaps re-checked against primary sources the same day. All claims are traced to primary sources (system cards, model cards, or named external evaluators). Where a finding could not be located, it is marked as a gap rather than guessed.
The user wants a straightforward quickstart to run headroom (this repo —
headroom-ai, a context-compression HTTP proxy that sits in front of an
LLM API and shrinks tool outputs/history/RAG chunks before they hit the
model) alongside their globally-installed rtk ("Rust Token Killer",
Homebrew binary at /opt/homebrew/bin/rtk, wired into Claude Code via a
PreToolUse hook in ~/.claude/settings.json that rewrites Bash commands
Both goldens are stopped, so nothing will collide. Here's where you stand.
Done (committed as 5a11722)
The prereq spec is fully executed and the branch is clean. just check is green — 328/328 claims, lint/typecheck/pyrefly/127 tests, packer fmt/init/validate.
Your next step: Phase B — the hero screenshot
This is the one piece nobody can automate. It needs you in the guest window.