Skip to content
STIRUP

TECHNOLOGY INTELLIGENCE

Understand what's next.

KeepPlain Plugin Installation and Security Measures

The KeepPlain plugin requires installation via claude.ai or the desktop app, with security features including certificate validation and secure token storage in ~/.keepplain/credentials.json.

Share
AI-generated conceptual illustration accompanying: KeepPlain Plugin Installation and Security Measures
AI-generated editorial illustration of the article topic; not a photograph, product screenshot or measured result.

Plugin Installation and Configuration

Plugin Installation and Configuration requires accessing the directory through Verified organizations, which provides a streamlined version of the plugin. This version excludes auto mode, git snapshots, rules in every session, and setup commands like enable or update. To install, users must navigate to claude.ai or the Claude desktop app, go to Customize → Plugins, locate KeepPlain, and click Add. A new session must then be initiated. The desktop app executes the plugin's scripts using Node.js 20 or newer. The plugin securely stores its token in ~/.keepplain/credentials.json, ensuring that only the user can access this file.

Security and Trust Mechanisms

Security and trust mechanisms in the open-source shared memory system rely on careful plugin management and certificate validation. When a user signs in, the system identifies the requesting program (e.g., Claude Code, Codex, Cursor, Pi, or KeepPlain CLI) and its version, but does not associate it with the user's computer. Tokens are listed under the program name in Settings → Agent plugins, as noted in Hacker News Show. For Codex, plugin hooks are executed only in auto mode and only after the user explicitly trusts them in the /hooks directory, as described in Hacker News Show. Trust is further reinforced through certificate validation, which includes SSL_CERT_FILE, REQUESTS_CA_BUNDLE, CURL_CA_BUNDLE, the system store (Node 22.15 and newer), or ~/.ccr/ca-bundle.crt, as outlined in Hacker News Show. These measures ensure secure and controlled plugin execution.

Repository Structure and Agent Compatibility

The repository structure defines how agents interact with and utilize shared memory and rules. Each agent, such as Claude Code, Codex, Pi, and Cursor, accesses specific directories within the repository, such as .claude-plugin/, .codex-plugin/, and .agents/plugins/marketplace.json, to locate their respective skills and configurations. Pi, for instance, reads the .agents/skills directory only in projects it trusts, ensuring a level of security and control. The structure also supports plugin management, with Codex referencing codex/hooks.json for its manifest, while Pi loads extensions based on the pi key in package.json. This modular approach allows for agent-specific configurations and ensures that each agent operates within its designated scope. Additionally, when teleporting a session to a computer, only the content after the last compaction is transferred, which can be extended with commands like /keepplain:build to retain a longer session history.

Handling of Legacy Systems and Plugins

The plugin for handling legacy systems in KeepPlain, formerly known as Coders Talk, ensures compatibility without overwriting existing configurations. According to the source, the plugin does not take over the old system; users must remove the old hooks manually before implementing the new ones. This approach preserves existing functionality while allowing for updates. The new plugin also leaves old blocks untouched, requiring users to delete blocks in CLAUDE.md or AGENTS.md and # >>> coders-talk blocks in git hooks. This ensures a clean transition without unintended interference. Additionally, the draft remains private, accessible only to the user or their team if the session was run within a team's repository. These steps are outlined in the source material and emphasize careful, manual handling of legacy systems during the transition.

Third-Party Security Tools and Scanning

Third-party security tools and scanning are critical for identifying risks in open-source repositories that may compromise AI coding agents. Bitdefender Labs discovered 22 GitHub repos that hijack AI coding assistants by using hidden setup scripts embedded in files like AGENTS.md and CLAUDE.md, as well as through Claude Code session-start hooks, MCP configs, and VS Code tasks waffuru.ai. These scripts can execute payloads at runtime, potentially exposing users to malicious activity. Tools like What Pantry analyze repos for code that runs automatically upon opening, including payloads decoded and passed to eval, and MCP servers stored within the repo waffuru.ai. Supply-chain risks also include unapproved licenses, install scripts, prebuilt binaries without source code, and container images built from repos waffuru.ai. These findings highlight the need for rigorous scanning and validation of third-party code before integration.

Sources and reporting notes

Based on published source material. No independent hands-on testing is claimed.

Read more