Claude Guardrails Skill

    1

    Guardrail pack: long-term memory discipline, fact verification, and a self-registering session-start reminder. Coordinates superpowers, planning-with-files, caveman, find-skills, and skill-judge.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    claude-guardrails-skill

    Example session with this skill installed

    I need to update the API client. I remember we use v2.4, but I should double check that and store our new timeout preference of 30 seconds for future sessions.

    • Read your context and instructions
    • Compiled the claude-guardrails-

    I have initiated the verification process for the library version and updated the long-term memory with your timeout preferences.

    Memory and Verification Audit

    ActionSubjectStatusResult/Location
    Fact VerifyLibrary VersionVerifiedConfirmed v2.4.1 via package.json
    Memory WriteTimeout PreferenceCreatedmemory/feedback-api-timeout.md
    Index UpdateMEMORY.mdUpdatedAdded pointer to timeout preference
    ---
    name: api-timeout-preference
    description: Global timeout setting for internal API clients
    metadata:
      type: feedback
    ---
    
    # API Timeout Preference
    
    **Why:** To prevent hanging connections on slow legacy endpoints during batch processing.
    **How to apply:** Ensure all instantiated API clients set the `timeout` parameter to `30000`ms.
    

    Next steps

    1. Review the memory/feedback-api-timeout.md file to ensure the "Why" and "How to apply" sections meet your requirements.
    2. Proceed with the API client update now that the v2.4.1 version is verified.
    3. Run verify-facts again if you need to confirm specific method signatures within the v2.4.1 SDK.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Managing context across long-running projects often leads to "memory rot" where agents hallucinate stale API signatures or clutter long-term storage with transient task data. This guardrail pack enforces strict discipline on how facts are stored, recalled, and verified to ensure your agent stays grounded in reality.

    What it does

    • Layered context management separates durable long-term facts from active task progress and implementation blueprints.
    • Atomic memory storage tracks user preferences, project constraints, and external references in single-fact files with metadata.
    • Fact verification intercepts factual claims about versions, APIs, or pricing to ensure they are confirmed in the current session.
    • Recall hygiene validates that memories naming specific files or functions are still accurate before they are used in a prompt.
    • Reference indexing maintains a clean MEMORY.md index that points to granular files rather than bloating the system prompt with body text.

    How it works

    1. Initialize the memory directory structure and MEMORY.md index at the start of a session.
    2. Commit durable facts by creating individual files with frontmatter for identity, feedback, or project goals.
    3. Verify any external claim—like a library version or config flag—by reading local source or fetching docs before stating it as fact.
    4. Update existing memory files when project constraints change, deleting or linking related facts via internal slugs.

    Frameworks & tools

    This skill is designed for agents with file-system access and web-search capabilities. It integrates with planning-with-files and superpowers workflows to manage state across Markdown-based documentation.

    Why this beats prompting it yourself

    Standard prompts often fail to distinguish between "what I am doing now" and "what is true about the project." This skill provides a structural framework that prevents task-specific noise from polluting the agent's long-term memory while mandating proof for factual assertions.

    Use cases

    • Tracking developer preferences and feedback across multiple coding sessions.
    • Maintaining project-specific constraints that are not visible in the git history.
    • Auditing API signatures and version-specific library behaviors before refactoring.
    • Organizing external reference links and dashboard URLs for easy retrieval.

    Known limitations

    Requires a specific file structure including a memory directory and index file. It does not automate the actual fetch or read calls but mandates that the agent perform them before asserting facts.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Requires file-system write access for memory storage. Relies on the agent having tools to fetch or read files to perform the verification steps.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    • Free with an Agensi account

    Software
    claude-guardrails-skillNode.jsargvdouble-fireGit

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables
    Listed19 days ago

    What's inside

    Frequently Asked Questions

    This skill prevents 'context rot' by enforcing a structured file-based system that separates permanent project rules from temporary task data, ensuring the agent doesn't hallucinate stale information during long-running projects.

    The skill is optimized for AI agents with file-system access and web-search capabilities, particularly those utilizing Markdown-based workflows like 'planning-with-files'.

    You will receive the complete logic for the memory directory structure, the MEMORY.md indexing system, and the specific behavioral guardrails that govern fact verification and storage.

    No, it does not use a vector database; instead, it uses a transparent, human-readable file system that allows you to audit, edit, and manage your agent's memories manually if needed.

    Setup involves initializing a specific memory directory and index file within your project workspace, which the agent then manages autonomously according to the provided guardrail rules.

    The skill mandates that an agent must perform a fresh 'read' or 'fetch' of a resource before asserting its state as a fact, preventing the use of outdated cached information from previous sessions.

    New in LLM Ops

    Free