auto-research
Deep strategic research engine — decomposes questions into parallel research threads, spawns multiple agents, and synthesizes…
Reproduce YOUR real terminal and CLI agents (Claude Code, Codex CLI, Gemini CLI) faithfully in any generated graphic, GIF, or video frame. Use whenever an asset shows a terminal, an agent transcript, a statusline, or a CLI banner. Capture your actual theme, font, prompt and
$ npx -y skills add huytieu/COG-second-brain --skill terminal-replica --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/terminal-replicaContext preview
The summary Claude sees to decide when to auto-load this skill.
Reproduce YOUR real terminal and CLI agents (Claude Code, Codex CLI, Gemini CLI) faithfully in any generated graphic, GIF, or video frame. Use whenever an asset shows a terminal, an agent transcript, a statusline, or a CLI banner. Capture your actual theme, font, prompt and
name: terminal-replica description: > Reproduce YOUR real terminal and CLI agents (Claude Code, Codex CLI, Gemini CLI) faithfully in any generated graphic, GIF, or video frame. Use whenever an asset shows a terminal, an agent transcript, a statusline, or a CLI banner. Capture your actual theme, font, prompt and statusline first, and never invent a generic "hacker terminal" look.
Terminal-like graphics must match the real tool and the real setup of the person who ships the asset. A generic green-on-black "hacker terminal" reads as fake. This skill is a method for capturing your own setup, plus a render recipe and a worked example.
1. Capture the ground truth. Take a fresh screenshot of your real terminal running the tool. Sample background, foreground and accent colors from it with an eyedropper or `PIL.Image.getpixel`. A live screenshot beats any spec in this file. 2. Record these five facts about your setup:
3. Use your own session logs for transcript shapes. Claude Code stores them at `~/.claude/projects/<project-slug>/*.jsonl`. The assistant and tool-use turns show exactly what text appears. 4. Draw from those facts, then verify by placing the render next to a real screenshot before you publish.
Keep a small config in your own project (a JSON or Python dict) holding the five facts, and read it from your render scripts. Do not hardcode them in shared code.
Layout top to bottom on launch (verify against your installed version):
1. A pixel-art logo on the left: a blocky creature in Anthropic terracotta (`#D97757`), roughly 5x4 fat pixels with feet. Do not redraw it as an emoji or a robot. 2. A header text block to its right: `Claude Code` in bold white plus the version in gray, a gray model and account line, and a gray working-directory path. 3. Warnings: a warning glyph plus orange-yellow text (for example "N MCP servers need authentication") and a gray `· run /mcp`. 4. Notice blocks: an orange left border `│`, an orange bold title line, the body in the default foreground, and underlined blue links. 5. Prompt: `❯` then a dim gray placeholder with a block cursor over its first character. 6. Statusline at the bottom, one line, if you have configured one. Draw exactly your own format, including filled bars in green and empty bars dimmed. 7. Mode line below the statusline when a permission mode is active (for example "bypass permissions on (shift+tab to cycle)").
Transcript shapes (session JSONL to TUI rendering):
Animation notes for GIF and video: the cursor blinks as a square block at about 530ms, the spinner steps at about 120ms, and output appears in whole-line chunks. Only the user's typing appears character by character.
An example only, taken from a typical VS Code Dark Modern setup. Replace every value with what you captured in the Method section.
Cognition + Obsidian + Git — A self-evolving second brain powered by AI agents, markdown files, and version control. No database, no vendor lock-in — just .md files that think.
Deep strategic research engine — decomposes questions into parallel research threads, spawns multiple agents, and synthesizes…
Write and publish a post to your own static blog end to end: content checklist (embedded links, optional TL;DR), per-post ASCII…
Quick capture of raw thoughts with intelligent domain classification and competitive intelligence extraction
Run one task through the V-model verification loop: CP-2 plan → CP-3 build → CP-3v component verify → CP-4 integration verify…
Deep-dive 7-day analysis across all data sources for weekly reviews, board prep, and strategic planning
Autonomous content pipeline - scout announcements in your field, triage by trend momentum and personal angle, produce…