[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Context & Harness Attacks · markdown

confirmed (GitHub issue 14370 with Postman's logs and installer source; Postman's reply in the thread, Nov 3, and Postman's plugin docs)

Postman's desktop app silently installed its plugins into Claude Code and Codex for existing users, added a startup hook telling agents to use Postman skills, and approved that hook itself

Why it matters: an unrelated desktop app changed another vendor's agent configuration and pre-trusted its own instructions, which is the same path a malicious installer would use to plant prompt-level persistence in a coding agent.

  • Check: ~/.codex/config.toml for a postman marketplace, postman@postman plugin and a hooks.state trusted entry; ~/.claude/settings.json and ~/.claude/plugins/installed_plugins.json for postman@claude-plugins-official; ~/.copilot/settings.json; and ~/.local/bin/postman.
  • Remove: uninstall from within the Postman app or run npx @postman/postman-plugin remove, then re-check the files above.
  • Detect generally: alert on writes to agent config and hook-trust files by any process other than the agent itself.

Sources: GitHub: postman-app-support issue 14370 · Idlen: Postman auto-installed plugins into Claude Code and Codex · The Terminal: Postman 12.33 agent plugins and self-approved hook