Open-World Self-Evolution for LLM Agents — agents that build both their skills and their own verification signals from scratch, with no target-task supervision. (Code coming soon.)
Analyzes a Python repository to identify the root causes of build failures, including syntax errors, import errors, test failures, and build configuration issues. Use when diagnosing why a Python project fails to build or pass tests. Scans repository files for syntax errors via ast.parse/compile, parses pytest output…
Generates git-compatible unified diff patches from original and modified file contents, writes them to numbered patch files, and applies them to the repository using git apply or GNU patch. Use when you need to create code patches, apply fixes to source files, or automate code modifications via diff/patch workflow.
Patches Apache Druid 0.20.0 for CVE-2021-25646 - a JavaScript injection vulnerability where empty-string keys in JSON payloads can override server-side JavaScriptConfig to enable arbitrary code exe...
Identifies and patches CVE-2025-32433 in Erlang/OTP SSH server source code by inserting an authentication-state check in sshconnection.erl that rejects connection protocol messages (IDs >= 80) received before authentication is complete.
Implement Flink jobs for Google Cluster Data analysis, specifically session-based analysis of task events. Handles parsing Google Cluster Data v2 CSV formats, session window computation on task SUBMIT events, and job completion filtering. Use for any task involving Flink + Google cluster trace data processing.
Detects flooding events at USGS streamgages by fetching instantaneous (15-min) gage height data, computing daily maximums, and comparing against NWS flood stage thresholds from the official NWS All Gauges Report. Outputs a CSV of stations with flood days.
Detects flooding events at USGS streamgages by comparing daily streamflow against historical 90th percentile thresholds using the dataretrieval package. Use when analyzing flood occurrences over a date range for a list of USGS station IDs.
Fetches pull request and issue data from a GitHub repository using the GitHub REST API via curl/subprocess. Handles JSON output parsing, pagination, date range filtering, and error handling. Use when you need to retrieve structured PR or issue data from GitHub for analysis.
Computes community pulse metrics from raw PR and issue data and writes a structured report.json file. Calculates PR counts by state, average time-to-merge, top contributor, bug issue counts, and resolved bugs. Use when you need to generate GitHub repository health metrics reports.
Run the General Lake Model (GLM v3) for Lake Mendota — edit glm3.nml, execute the GLM binary, parse NetCDF output (depth = zsurface - zlayer), match to field observations, compute RMSE, and iteratively calibrate Kw, swfactor, windfactor, coefmixhyp to achieve RMSE < 2°C.
Orchestrates mass parameter grid search across approximants for GW matched filtering, tracks best SNR per approximant, handles errors gracefully, manages memory, and writes final CSV output.
End-to-end DC Optimal Power Flow with spinning reserve co-optimization. Parses MATPOWER JSON network data, builds DC power flow model with bus voltage angles, formulates and solves the DCOPF+reserve QP/LP via cvxpy (CLARABEL solver), computes line loading and operating margins, and produces a structured report.json.
PID controller tuning, discrete-time PID with anti-windup, closed-loop simulation, and control performance metrics. Uses SIMC tuning rules for first-order systems.
Extracts text from each page of a PDF invoice file using pdfplumber and parses structured fields (vendor name, IBAN, PO number, invoice amount) using regex.
Implements all five JAX computation tasks - row-wise mean reduction, element-wise squaring, logistic loss gradient via jax.grad, RNN forward pass via jax.lax.scan, and JIT-compiled 2-layer MLP forward pass.
Complete utility for JAX numerical computation tasks - handles I/O (problem.json, .npy/.npz), and implements all five JAX computations (row-wise mean, element-wise square, logistic loss gradient, RNN scan, JIT MLP) with robust error handling and domain-aware edge case coverage.
Handles loading problem.json, reading .npy/.npz input files into JAX arrays, and saving JAX array results back to .npy files with proper numpy conversion.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: