Claude Skill MartBrowse skillsQuick linesLearn by videoTerminal guideWhat is a Skill?
 
← Back to list

Agent Context Isolation

Rules that keep subagent transcripts out of your main context by running agents in the background and passing results through files.

AutomationIntermediate★ 530⑂ 44AI score 7/10Last updated: Aug 8, 2026

What it does

  • Enforces launching subagents with run_in_background=true so their transcripts stay isolated from the main conversation.
  • Replaces TaskOutput (which dumps 70k+ token transcripts) with file-based coordination: agents write to .claude/cache/agents/<agent-type>/, the main session just reads those files.
  • Gives a concrete completion / stuck-agent detection procedure using progress system reminders, output-file polling, and output size growth.
  • Requires verification through real test runs (bun test) and pass/fail counts instead of trusting the agent's narrative.
  • Provides a file-based agent pipeline pattern: research → plan → validate → implement, each stage reading the previous stage's file.

Who it's for

  • Claude Code users who've hit mid-session compaction after fanning out multiple subagents.
  • Anyone orchestrating staged agents (research, planning, implementation) on larger codebases.
  • Developers who want to stretch their token budget across long sessions.

Examples

  1. Send a large refactor investigation to a background research agent, then read only .claude/cache/agents/oracle/output.md in the main session to summarize.
  2. Build a four-stage pipeline where the plan agent's file is validated by another agent before the implementation agent touches source files.
  3. When progress reminders stop, run find .claude/cache/agents -name "*.md" -mmin -5 to tell a finished agent from a stuck one.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/vibeeval/vibecosystem/HEAD/skills/agent-context-isolation/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on Claude Skill Mart.
Copy the skills/agent-context-isolation folder from the GitHub repo vibeeval/vibecosystem into my ~/.claude/skills/vibeeval-agent-context-isolation/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/vibeeval/vibecosystem.git && mkdir -p ~/.claude/skills && cp -r vibecosystem/skills/agent-context-isolation ~/.claude/skills/

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open a terminal.
  2. Clone the repository: git clone https://github.com/vibeeval/vibecosystem.git
  3. Create the skills folder if needed: mkdir -p ~/.claude/skills
  4. Copy the skill in: cp -r vibecosystem/skills/agent-context-isolation ~/.claude/skills/
  5. Restart Claude Code and confirm the skill is loaded. (It sets user-invocable: false, so it acts as a reference ruleset during agent orchestration rather than a command you invoke.)
  6. Adapt it to your project: swap the test command (bun test → npm test, pytest, etc.) and adjust the .claude/cache/agents/... output paths.
View source on GitHub ↗License: MIT