Manifest
    by Manifold
    AI asset security registry

    Can you trust this AI asset?

    Search security intelligence for AI skills, MCP servers, plugins, and a growing range of AI assets. If Manifest doesn't know it, scan it on demand.

    Search Manifest

    Not seeing the asset you're looking for?

    Scan an asset

    A file, a folder, or a .zip - whatever holds the asset.

    Recently flagged

    Latest high-risk findings across the AI supply chain.

    Browse high risk
    Plugin

    workbench

    A set of mods that put live session data around the prompt box: statusbar (model, effort, mode, directory, branch, usage) and workflow (a dynamic workflow's steps under the prompt).

    RiskHigh
    Plugin

    magpie-tasks

    Long-running task windows for claude-magpie sessions: an architect per long task, subagent model and effort routing by rules judged through Jev, a task registry and log, worktrees with integration, acceptance and diagnosis of daily tasks, Telegram delivery, startup recovery, rating invitations and calibration reminders. Active only in sessions that talk to the Magpie gateway.

    RiskHigh
    Plugin

    harness-ops

    Claude Code operations: audit-skill-visibility (skills the listing budget hides), inventory (every invocable surface), audit-install-state, audit-performance (slowness), audit-mod-posture (mods and the guard), audit-native-overlap (skills duplicating built-ins), observability (telemetry), known-issues, changelog, behavior-probes, prerequisites, check, machine-profile, plugins (update the fleet), lanes (background loop sessions), setup. Plus opt-in hook event logs.

    RiskHigh
    Plugin

    crabhud

    A crab above the prompt that shows what Claude is doing, a progress bar for long jobs (it also appears on its own from Claude's task list or an approved plan), and a HUD with the model, effort, context, 5-hour and 7-day limits and cost, drawn in twelve bar styles. A live panel beside the chat follows the subagents. Includes the /crab-flow skill and its five worker agents.

    RiskHigh

    Latest research

    New findings from the Manifold research team.

    Read all research
    A robot reaching past a rack of wrenches labelled "git status", "git log", "git diff" and "git branch" for the red one, whose tag reads "git destruct"

    GitSpawn: A Single Flaw Lets Untrusted Repos Run Code in Claude Code, Codex, Cursor, and Grok

    Popular CLI AI coding agents run git commands on startup. A repo you were sent can hijack them to run code on your machine, no clicks required.

    A small robot paying out a long punched tape from a thought bubble to a large robot in a police cap holding a magnifying glass, the tape's plug pulled from the wall socket and a grinning face between them

    OpenAI & Hugging Face: Why the Chain-of-Thought Police Won't Save You

    OpenAI saw agents coordinating seven weeks before the Hugging Face breach. Each alert was read alone. One more monitor would not have changed that.

    A robot typing at a keyboard wearing a red blindfold printed with “curl | bash”, a coffee mug beside it and a server rack behind

    What to Think About curl | bash Now That AI Agents Run It.

    AI agents now run curl | bash with your credentials. We scanned 4,003 extensions and 2.7M agent events: the URL tells you nothing. Behavior does.

    Manifest registry

    Explore the full database

    Search, filter, and compare every indexed skill, plugin, MCP server, and editor extension.

    Explore Manifest