loopsmith

    1

    Learn from coding-agent traces and safely promote useful lessons into versioned instructions.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    loopsmith

    Example session with this skill installed

    Import the trace at ./traces/auth_fix.jsonl for the 'auth-service' project and mine it for improvements.

    • Read your context and instructions
    • Compiled the loopsmith

    I have imported the session trace and mined it for new instructions. LoopSmith identified one high-confidence proposal based on the agent's repeated correction of the JWT implementation during the session.

    LoopSmith Proposal: auth-service (ID: prop_82f1)

    FieldDetail
    Projectauth-service
    Source Traceauth_fix.jsonl
    Evidence Count3 citations
    StatusPending Review

    Proposed Change

    
    ## Authentication Patterns
    - Use standard library for JWT headers.
    + Always use the `PyJWT` library with explicit `algorithms=['RS256']` for token decoding to prevent algorithm confusion attacks.
    

    Cited Evidence

    1. Event #42: Agent manually corrected jwt.decode call after validation error.
    2. Event #89: Agent added algorithms parameter to fix runtime Warning.
    3. Event #112: Final successful test pass after explicit algorithm definition.

    Next steps

    1. Run loopsmith proposals show prop_82f1 to see the full context.
    2. Execute loopsmith eval prop_82f1 to run behavioral benchmarks against this change.
    3. Use loopsmith approve prop_82f1 if the evaluation verdict is promote.

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

    About this skill

    LoopSmith turns coding-session history into reusable project guidance—and checks proposed changes before they become active.

    Import the JSONL traces you choose. LoopSmith mines them for lessons and creates proposals that cite the trace events they came from, so you can inspect the evidence and proposed wording. Deterministic mining works without a model. If you configure an OpenAI-compatible endpoint, LoopSmith can also help mine proposals and compare current guidance with a candidate version on benchmark tasks.

    Static checks look for invalid proposals, unsupported evidence, stale targets, and changes that do nothing. Behavioral evaluation runs a baseline and candidate in isolated task workspaces and grades them against test checkers. A candidate is promoted only after evaluation and your approval. Each approved change is versioned, auditable, reversible, and exportable as AGENTS.md, memory, or agent skills.

    LoopSmith works in batches from imported traces; it does not watch live sessions. Model-assisted mining and behavioral evaluation send relevant content to the endpoint you configure and may use API credits. Benchmark results describe performance on the included scenarios, not a guarantee of improvement on every coding task.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Requires local LoopSmith CLI installation. No live session monitoring. Evaluation requires OpenAI-compatible API keys for behavioral benchmarks.

    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 6 days ago

    • Free with an Agensi account

    Accounts
    GitHubExample
    Software
    Python

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables

    Allowed Hosts

    github.com
    example.test
    Listed8 days ago
    Updated6 days ago

    What's inside

    Frequently Asked Questions

    LoopSmith analyzes your agent's JSONL coding logs to identify exactly where errors occurred and generates evidence-backed instruction updates to prevent those specific failures in the future. It replaces trial-and-error prompting with a data-driven process that cites specific trace events for every change.

    This skill is designed to work with any coding agent that can export JSONL session traces and supports instructions via Markdown or memory files. It uses OpenAI-compatible endpoints for its internal mining and evaluation, making it flexible across different LLM providers.

    No, LoopSmith operates in an offline, batch-processing mode. You import completed session logs into its local SQLite database to analyze performance and benchmark new instructions before deploying them to your live agent.

    The 'Evaluate' step runs your candidate instructions against a suite of benchmark tasks using an OpenAI-compatible API to verify improvements. This ensures that new instructions actually fix the targeted issue without causing regressions in other areas of the agent's capabilities.

    Setting up requires a local Python environment and a source checkout of the LoopSmith package. Once installed, you simply run `loopsmith init` in your project directory to initialize the local SQLite database used for trace storage and analysis.

    New in Agents & Orchestration

    Free