the-decoder.com web signal

Anthropic Ships 'Record a Skill' in Claude Cowork Desktop App

5 sources tracking this story

TL;DR

  • OpenAI launched a comparable 'Record and Replay' feature in June, making demonstration-based skill capture an industry standard within a single product cycle.
  • AlphaSignal confirms Record a Skill replaces the previous SKILL.md manual-authoring requirement, cutting skill creation from technical documentation to narrated screen recording.
  • CyberSecurity News specifies access is gated to Pro, Max, and Team plans only, excluding Free tier, establishing the feature as an explicit upsell lever.

Anthropic dropped a small workflow-automation feature into Claude Cowork this week that reads like a nudge toward how non-engineers might actually get value out of agent-style products. Starting July 21, 2026, Pro, Max, and Team users can hit 'Record a skill' in the + menu of the Claude desktop app, screen-record themselves doing a task while talking through it, and get back a skill Claude can run again, according to The Decoder.

The pitch is straightforward: don't write instructions, just do the thing. In a post on X, Claude's account described it as 'record your screen while you do a task, talk through it as you go, and Claude turns it into a skill it can run again.' That framing matters because the group most likely to automate their own work, operations, finance, support, back-office, is also the group least likely to sit down and hand-write clean agent prompts.

It is also a competitive move. The Decoder notes that OpenAI offers a similar feature in its Codex tool, so what is really being tested is whose desktop client owns the surface where knowledge workers demonstrate their tasks. The Team-plan availability is the interesting tell there. A shared library of company-specific skills is a stickier moat than any single benchmark win, and it is the kind of surface area a rival cannot easily copy just by matching a model score.

The honest caveat is what the reporting doesn't cover. There is no detail on what happens when the underlying app UI changes and a replay breaks, how the recording handles credentials or sensitive data that scroll past on screen, or whether skills can be shared and edited across a Team plan's users. Those are the questions that decide whether this becomes real infrastructure inside a business or a demo people try once and forget.

The upside, if it holds up on messy real-world workflows, is that skill authoring shifts from a prompt-engineering specialty back to something a domain expert can produce in a few minutes. That is the direction worth watching, even before the feature has proven itself outside the launch clip.

What others are reporting

Coverage cluster as of 24h after publish

  1. Android Authority Read →

    Frames the feature as solving the AI re-explanation problem; includes live rollout details and where to locate it in Claude Pro.

    Instead of typing out long instructions, you simply record your screen while completing a task yourself and explain what you're doing as you go.
  2. Search Engine Journal Read →

    Adds competitive context citing OpenAI's Record and Replay from June; documents early public concern about job replacement that Anthropic's rollout messaging does not address.

  3. AlphaSignal Read →

    Contrasts the new demo-based method with the previous SKILL.md manual-authoring requirement; distinguishes individual skills from bundled plugins.

    Hit record, do the task on your screen while narrating your reasoning out loud, stop the recording, and Claude synthesizes everything into a skill it can replay later.
  4. CyberSecurity News Read →

    Specifies plan-tier gating (Pro, Max, Team only; Free excluded) and frames the capability as an 'AI employee' that learns from observed behavior.

    Record your screen while you do a task, talk through it as you go, and Claude turns it into a skill it can run again.