- Home
- Skills
- Agents & Orchestration
- Autonomous Coder Agent
Autonomous Coder Agent
A local agentic software repair engine using Ollama and LanceDB to diagnose, repair, and verify codebases via a closed-loop ReAct system.
Free
Works with the AI Agents you already use
Autonomous Coder Agent
Example session with this skill installed
There is a bug in the discount calculation logic. It seems to be applying the discount twice. Please use search_codebase to find the logic, patch the bug, and run the tests to verify the fix.
- Read your context and instructions
- Compiled the autonomous coder agent
Summary
I successfully located, patched, and verified the double-discount calculation bug in src/billing.py using the local ReAct loop and execution sandbox.
| Metric | Before Patch | After Patch |
|---|---|---|
Test Results (test_billing.py) | 1 Failed, 2 Passed | 3 Passed, 0 Failed |
| Code State | Double deduction | Single deduction |
# The corrected logic applied to src/billing.py
def calculate_discount(price, discount_rate):
"""
Calculates the final price after applying the discount rate.
"""
# Fixed: Removed the redundant secondary subtraction that caused the bug
return price - (price * discount_rate)
Connects securely to your tools. The creator never sees your data.
About this skill
Software repair is often slowed down by manual context switching and slow feedback loops. This skill delivers an autonomous, local-first engine that diagnoses bugs and applies verified patches within your environment.
What it does
- Semantic Code Search indexes your codebase using LanceDB and HuggingFace embeddings to find relevant files by architectural intent.
- Automated Patching applies structural code modifications directly to local workspace files based on identified issues.
- Closed-Loop Verification runs
pytestsuites via subprocess and feeds terminal logs back into the reasoning engine. - Failure Mitigation uses a specialized JSON brace parser to handle schema inconsistencies from smaller open-weights models.
- Iterative Debugging implements a ReAct loop to observe test results and re-attempt repairs until the suite passes.
How it works
- Index the codebase into a local vector store to enable RAG-based context retrieval.
- Search for the specific code modules or functions relevant to the reported bug or feature request.
- Execute a repair cycle where the agent reads the code, applies a patch, and runs the test harness.
- Iterate automatically if tests fail, using the console output to refine the logic.
Frameworks & tools
Built for local environments using Ollama, LanceDB, and HuggingFace (all-MiniLM-L6-v2). It integrates with Python projects using pytest for validation.
Why this beats prompting it yourself
Prompting alone lacks the stateful loop required to verify if a suggested fix actually runs. This skill bridges the gap between LLM reasoning and OS-level execution, ensuring code is not just written, but validated against your existing test suite.
Use cases
- Fix failing unit tests in local Python repositories.
- Refactor legacy code by searching for patterns and applying verified updates.
- Scaffold and test new feature implementations without manual terminal commands.
- Resolve architectural inconsistencies across large codebases using semantic search.
Known limitations
Requires a local Ollama instance and Python environment with pytest installed. Best suited for qwen2.5-coder or similar models capable of structured JSON output.
Requires local Ollama and LanceDB setup. Designed specifically for Python codebases using pytest. Performance depends on the local model's ability to follow JSON schemas.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
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
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 14 days ago
- Free with an Agensi account
Requires
Permissions
Requires Ollama running locally with the qwen2.5-coder:7b model pulled prior to execution.