Ted S (Ted)

[Ted S (Ted) @ted_sandico]

anthropics/skills by Anthropic: +4,017 GitHub stars this week.

Skills are folders of instructions, scripts and resources Claude loads for specialized tasks. In the PDF skill, a script reads a form's fields and writes them as JSON.

Source: GitHub, anthropics/skills, 2026-11-11

[Ted S (Ted) @ted_sandico]

Denied is not done: check what ran after a refused call.

New in Claude Code: permissions.denyInsteadOfPrompting. A call that would wait on a permission prompt is denied and the turn continues, subagents too.

Source: Claude Code changelog 2.1.319. Not affiliated with Anthropic.

[Ted S (Ted) @ted_sandico]

Stop asking AI for motion graphics. Ask it for code.
Code states every frame, so the text comes out spelled right and a fix is an edit, not a re-roll.

[Ted S (Ted) @ted_sandico]

Did you know your Antigravity agent reads skills you may never have looked at?

/skills lists every loaded skill, local and global, with its name and description.

We ran it in an empty folder: fifteen came back. Check what's steering your agent.

[Ted S (Ted) @ted_sandico]

The call you missed at 9 PM might have been your best lead this week.

ElevenLabs Agents answer inbound calls, book the meeting on your calendar, and save the caller to your CRM.

Flow inspired by Damola. Not affiliated, just a happy user.

[Ted S (Ted) @ted_sandico]

Taste Skill gets coding agents to design websites instead of the average landing page. 3,822 stars this week.

Start at skills/taste-skill/SKILL.md. Every page gets checked in a real browser.

Repo: Leonxlnx/taste-skill by @Leonxlnx
https://t.co/dt73pTVvVW

[Ted S (Ted) @ted_sandico]

Our Claude Code setup: 31 allow rules, 0 deny.

Our one hard rule lived in a hook that missed gh pr merge.

/permissions gives you Allow, Ask and Deny lists Claude checks on every tool call. One deny rule closed the gap.

[Ted S (Ted) @ted_sandico]

obra/superpowers turns coding agents into a full dev workflow built from composable skills: brainstorm, plan, TDD, review, then ship. 1,878 stars this week.

Repo: obra/superpowers by @obra
https://t.co/mjtTUxGUZZ

[Ted S (Ted) @ted_sandico]

You switch Claude Code to a lighter model for one quick fix, and it quietly becomes your default.

Type /model, pick the model, press s. That session only. Your default stays put.

Enter is the one that changes your default.

[Ted S (Ted) @ted_sandico]

DeepSeek Harness: open-source agent harness from DeepSeek AI, 1,467 stars this week. Every tool call goes through approval, and allowed-once is the only grant.

Repo: deepseek-ai/deepseek-harness by @deepseek-ai
https://t.co/jMghMk9SaG

[Ted S (Ted) @ted_sandico]

Claude stops when it thinks it's done. /goal makes a second model check.

You set a finish line. After every turn, a small, fast model reads the chat and rules: met, not yet, or impossible. Not yet, Claude keeps going.

It can't run anything itself, so make Claude print the proof.

What finish line would you hand it first?

[Ted S (Ted) @ted_sandico]

Claude Code shows you exactly what's filling its memory: /context

We ran it in a fresh session before typing a word. 4% of the window was already spoken for. Biggest item: our AGENTS file, at 3,700 tokens.

It only looks. Nothing gets deleted.

#ClaudeCode

[Ted S (Ted) @ted_sandico]

Claude Code can bill 3 ways: your Claude subscription, API usage via a Console account, or a platform like Amazon Bedrock.

/login picks one. /status shows which one you're on.

Our server still said Pro. We pay for Max. One /login fixed it.

[Ted S (Ted) @ted_sandico]

Switching tasks in Claude Code? Run /clear.

We tested it: context dropped from 79.2k to 39.6k tokens (8% → 4% of the window). Messages went from 41.1k to 1.6k. Project memory stayed loaded.

Label it (/clear my-task) and the old chat waits in /resume.

[Ted S (Ted) @ted_sandico]

Did you know Antigravity shows how much model quota you have left?

/usage (or /quota) opens the Model Quotas panel: each pool's weekly and five-hour limit, plus the reset time.

We ran it headless: Gemini at 97%, Claude and GPT at 35%. Check before a long task.

[Ted S (Ted) @ted_sandico]

@jonasmassieai Flow inspired by these two builds, worth watching both:
→ Alex Followell: https://t.co/4mIhLsXzoh
→ Jonas Massie (@jonasmassieai): https://t.co/NAZSuanOeI

[Ted S (Ted) @ted_sandico]

Your lead filled out a form 4 minutes ago. An AI agent would have booked the meeting already.

ElevenLabs Agents call every new lead in seconds, qualify them, and book the meeting.

Flow inspired by Alex Followell and @jonasmassieai. Not affiliated, just a happy user.

[Ted S (Ted) @ted_sandico]

One MCP for reverse engineering. rea binds Hopper, Ghidra, IDA or Kuna to one Binary session and returns every observation as Evidence. Start reading at scripts/rea.mjs.
Repo: morluto/rea by @morluto
https://t.co/Axkpcjpsbe

[Ted S (Ted) @ted_sandico]

One rule can stop Claude Code from ever running a command.

/permissions shows your allow, ask and deny lists.

Ours had 31 allow, 0 ask, 0 deny, so nothing stopped gh pr merge.

One deny rule later: 31 / 0 / 1.

If a rule isn't on the deny list, Claude can't know it.

[Ted S (Ted) @ted_sandico]

Claude Code tip: if you press Enter in the /model picker, your pick becomes the default for every new session.
Press s instead and the switch covers this session only.
We tested it: Opus 5.5 → Sonnet 5 at xhigh, pressed s. New sessions still default to Opus 5.5.