AWS automation AI agent asap 3 days to do

# AWS AgentCore Automation for S3-to-S3 File Copy ## Timeline The project must be completed within three days. must simulate scenarios when agents is needed for example artificial delay to simulate error in files coping The freelancer should develop and test the solution using their own AWS account. Alternatively, we can meet online, and the freelancer can guide me step by step while I configure and test the solution in my AWS account. ## Project Goal Create an educational AWS AgentCore automation system that automatically copies files from one Amazon S3 bucket to another Amazon S3 bucket. The automation must run automatically. It must not wait for a user to submit a natural-language request. The project must teach me how practical agentic automation can be implemented on AWS. The solution must use: * Amazon Bedrock AgentCore * Amazon S3 * AWS Lambda * AWS Step Functions * AWS IAM * Amazon CloudWatch Logs * Python * Boto3 ## Important Project Requirement This must be a real AWS AgentCore project, not only an S3 copy script. AWS Lambda and AWS Step Functions should perform deterministic operations. AWS AgentCore must observe the workflow, interpret AWS results and logs, make decisions, select actions, verify results, create reports, send alarms, and request human approval when necessary. ## Required Architecture The solution must include: * A source Amazon S3 bucket * A destination Amazon S3 bucket * An automatic trigger or schedule * AWS Step Functions for workflow orchestration * AWS Lambda functions for deterministic operations * AWS AgentCore for agentic monitoring and decision-making * Amazon CloudWatch Logs for monitoring * AWS IAM with least-privilege permissions * Reporting * Alarms * Human-in-the-loop approval The freelancer must explain: * Where the AgentCore agent runs * Where the agent loop runs * How AgentCore interacts with Step Functions * How AgentCore invokes Lambda functions * How AgentCore reads AWS results and CloudWatch logs * How workflow state is stored * How previous attempts are remembered * How the agent decides what to do next * How human approval is requested * Which functions require AI reasoning * Which functions should remain deterministic code ## Automatic Workflow The workflow should start automatically when: * A new file is added to the source S3 bucket * A scheduled copy process starts * A group of files is ready for processing The workflow must: 1. Detect files that need to be copied. 2. Inspect the source objects. 3. Start the Step Functions workflow. 4. Use Lambda functions to copy the files. 5. Copy the files to the destination S3 bucket. 6. Verify every copied file. 7. Read AWS execution results and CloudWatch logs. 8. Detect errors, warnings, delays, and unexpected results. 9. Retry temporary failures. 10. Stop retrying permanent failures. 11. Recover from partial batch failures. 12. Create a final report. 13. Send an alarm when human attention is required. 14. Request human approval before important or unsafe actions. ## Work the AWS AgentCore Agent Must Automate ### Eyeballs Work The agent must monitor: * S3 objects * Lambda results * Step Functions executions * CloudWatch logs * Errors * Warnings * Failed operations * Delayed operations * Verification results ### Thinking Work The agent must interpret the available evidence and decide whether to: * Continue * Retry * Skip * Stop * Copy a file again * Wait * Investigate another AWS result * Create an alarm * Request human approval * Escalate the problem ### Typing and Clicking Work The agent should perform approved AWS actions instead of requiring a person to repeatedly use AWS consoles or commands. ### Report Creation The agent must create reports showing: * Files discovered * Files copied * Files verified * Files skipped * Files that failed * Retry attempts * Verification results * Agent decisions * Human decisions * Errors and warnings * Final workflow status ### Alarms The agent must send an alarm when: * A file repeatedly fails * Verification fails * IAM permission is missing * A Lambda function repeatedly fails * A Step Functions execution fails * A workflow becomes unusually delayed * The destination bucket is unavailable * Conflicting AWS results are detected * Human attention is required ### Human in the Loop The agent must request human approval before important actions such as: * Overwriting a different destination object * Deleting an incorrect destination object * Restarting a large failed batch * Continuing after repeated verification failures * Making any security-sensitive change * Taking an action when the safe decision is uncertain The agent must not automatically grant itself broader IAM permissions. ## File Verification The system must verify that every copied file is correct. For every file, the solution must: 1. Calculate or retrieve the source checksum. 2. Record the source file size and relevant metadata. 3. Copy the file to the destination S3 bucket. 4. Calculate or retrieve the destination checksum. 5. Compare the source and destination checksums. 6. Compare the source and destination file sizes. 7. Report success only when verification passes. SHA-256 should be used where practical. The documentation must explain why an S3 ETag is not always a reliable integrity checksum. A copied file must not be reported as successful when verification fails. ## Freelancer Responsibility for Agent Functionality The scenarios listed in this job description are only examples. I am not providing a complete specification of every function the agent should have. The freelancer is responsible for designing the complete AWS AgentCore functionality required for a reliable agentic S3-to-S3 automation system. The freelancer must identify and implement all necessary agent functions, including functions not explicitly listed in this job description. The freelancer must determine: * What the agent needs to observe * Which AWS tools the agent needs * Which decisions the agent must make * Which failures must be recognized * Which errors are temporary * Which errors are permanent * Which actions can be automatic * Which actions require human approval * How the agent verifies that its action solved the problem * How the agent prevents infinite loops * How the agent prevents unsafe actions * How the agent reports its decisions The freelancer must not limit the implementation only to the example scenarios provided by me. ## Required Agent Tools The freelancer should design and implement all necessary AgentCore tools. The tools should include, where appropriate: * List source S3 objects * Read S3 object metadata * Read object size * Read or calculate object checksum * Check whether a destination object exists * Copy an S3 object * Verify a copied object * Read Step Functions execution status * Read Step Functions execution history * Read Lambda results * Read CloudWatch logs * Retry a failed operation * Stop a workflow safely * Resume a partial batch * Create a report * Create an alarm * Request human approval * Record the final decision The freelancer may add additional tools required for a complete and reliable implementation. ## Testing Requirements The freelancer must test every implemented agent function. The test scenarios listed below are examples only. The freelancer must develop a complete test plan covering all AgentCore functionality, AWS integrations, tools, decisions, recovery actions, alarms, reports, and human approval flows. Every agent tool and every important decision path must have at least one test. The freelancer must test: * Normal successful operation * Temporary failures * Permanent failures * Incorrect AWS configuration * Conflicting information * Partial completion * Repeated failures * Unsafe actions * Human approval * Recovery after failure * Verification after recovery * Reporting and alarms The freelancer must provide a test coverage table mapping: * Agent functionality * Agent tool * AWS service * Test scenario * Expected agent decision * Expected AWS action * Expected verification * Actual result * Pass or fail status ## Minimum Example Agent Scenarios The complete test plan must include at least 10 agent decision scenarios. These are minimum examples. The freelancer must add more tests when needed to cover all developed functionality. ### Scenario 1: Temporary AWS Failure Simulate a temporary AWS service or connection failure. The agent must: * Read the error * Recognize that it may be temporary * Retry using a limited retry policy * Verify the result after retry * Report the final outcome ### Scenario 2: Permanent IAM Failure Remove a required IAM permission. The agent must: * Detect the Access Denied error * Identify the failed AWS action * Stop unnecessary retries * Create an alarm * Explain which permission may be missing * Request human action ### Scenario 3: Verification Failure Simulate a destination checksum that does not match the source checksum. The agent must: * Detect the mismatch * Mark the copy as unverified * Investigate the logs and results * Decide whether to copy the file again * Perform a limited retry * Verify the new copy * Escalate if the mismatch continues ### Scenario 4: Partial Batch Failure Simulate a batch where some files succeed and other files fail. The agent must: * Separate successful files from failed files * Avoid copying verified files again * Investigate the failed files * Retry only the failed files * Verify the recovery * Create an updated batch report ### Scenario 5: Destination Object Already Exists Place an object with the same name in the destination bucket. The agent must inspect: * Source and destination sizes * Checksums * Metadata * Overwrite policy The agent must decide whether to: * Skip an identical object * Request approval to overwrite a different object * Preserve both objects * Report a conflict ### Scenario 6: Repeated Lambda Failure Simulate a Lambda function that fails repeatedly. The agent must: * Read Lambda and Step Functions results * Determine whether the failure is temporary or permanent * Stop after the retry limit * Avoid an infinite retry loop * Create an alarm * Provide a useful failure report ### Scenario 7: Delayed Step Functions Execution Simulate a workflow that remains in progress longer than expected. The agent must: * Detect the delay * Inspect execution history * Determine whether the workflow is still progressing * Decide whether to wait, stop, or escalate * Notify a human when the safe action is uncertain ### Scenario 8: Source Object Changes During Processing Replace or modify a source object during processing. The agent must: * Detect changed size, checksum, metadata, or version * Recognize that the original plan may no longer be valid * Stop verification of the previous version * Decide whether to restart the copy * Request human approval when required * Record the object version processed ### Scenario 9: Conflicting AWS Evidence Simulate a copy operation that reports success while verification data indicates a problem. The agent must: * Compare evidence from multiple AWS services * Avoid trusting only one success result * Identify the conflict * Mark the operation as unverified * Investigate the likely cause * Select a safe recovery action ### Scenario 10: Human Approval Required Simulate a situation where the agent cannot safely continue independently. The agent must: * Pause the workflow * Explain the situation * Present the available options * Request human approval * Continue only after approval * Record the human decision ## Evidence Required for Every Test For every test scenario, the freelancer must provide: * Functionality being tested * Purpose of the test * Method used to simulate the situation * AWS services involved * Initial agent observation * Relevant AWS results or CloudWatch logs * Agent classification * Agent decision * Tool selected by the agent * AWS action performed * Verification result * Human approval or escalation, where applicable * Expected result * Actual result * Screenshot or video evidence * Pass or fail status ## Required Deliverables ### Complete Working AWS AgentCore Project Provide one complete and working AWS AgentCore implementation. ### Complete Source Code The source code must include: * AgentCore agent code * AgentCore tools * Lambda functions * Step Functions definition * Python and Boto3 code * IAM policies * Configuration files * Deployment scripts or commands * Automatic trigger configuration * Logging * Retry logic * Verification logic * Reporting * Alarms * Human approval workflow * Test automation * Cleanup instructions No passwords, AWS access keys, or secrets may be hardcoded. ### Step-by-Step Instructional Video The video must show: * AWS account preparation * Source and destination bucket creation * IAM configuration * Lambda configuration * Step Functions configuration * CloudWatch configuration * AgentCore configuration * AgentCore deployment * Automatic workflow execution * Agent monitoring * Agent decisions * File copy * File verification * Failure simulation * Recovery * Human approval * Reports * Alarms * Test execution * AWS resource cleanup ### Word Document Provide a Word document with step-by-step instructions and screenshots for every major step. The document must include: * Architecture diagram * Windows prerequisites * AWS account preparation * AWS Console screenshots * S3 configuration * IAM configuration * Lambda configuration * Step Functions configuration * CloudWatch configuration * AgentCore configuration * AgentCore tools * Deployment instructions * Windows commands * Test setup * Complete test plan * Test evidence * Troubleshooting * Security recommendations * AWS cost considerations * Cleanup instructions ## Acceptance Criteria The project will be accepted only when: 1. The complete solution uses AWS AgentCore. 2. The project works in my own AWS account. 3. I can deploy, configure, run, monitor, and test the solution from my Windows laptop. 4. The documentation includes all required Windows commands. 5. Files are copied automatically from one S3 bucket to another S3 bucket. 6. No user natural-language request is required to start the copy process. 7. AWS Step Functions controls the workflow. 8. AWS Lambda performs deterministic file operations. 9. AWS AgentCore observes AWS services and reads execution results and logs. 10. AWS AgentCore makes different decisions for different situations. 11. The agent can continue, retry, skip, stop, wait, create an alarm, or request human approval. 12. Every copied file is verified. 13. Source and destination checksums are compared. 14. Source and destination file sizes are compared. 15. A file is reported as successful only after verification passes. 16. Temporary failures are retried using a limited retry policy. 17. Permanent failures are not continuously retried. 18. Infinite agent loops are prevented. 19. Partial batch failures can be recovered without copying all files again. 20. Successfully verified files are not copied again unnecessarily. 21. CloudWatch Logs show workflow actions, agent observations, decisions, retries, and results. 22. The agent creates a final report. 23. The agent creates alarms when human attention is required. 24. Human approval is demonstrated for at least one important action. 25. At least 10 agent decision scenarios are demonstrated. 26. The freelancer adds all additional tests required to cover the complete solution. 27. Every developed agent function is tested. 28. Every AgentCore tool is tested. 29. Every important agent decision path is tested. 30. A test coverage table is provided. 31. The instructional video demonstrates the complete process. 32. The Word document contains screenshots for every major AWS step. 33. I can run all tests myself from my Windows laptop. 34. I can reproduce the project in my AWS account without additional paid assistance. ## Final Acceptance Principle The example scenarios supplied by me are only starting examples. The freelancer is responsible for designing the complete AWS AgentCore solution, identifying all required agent functionalities, implementing those functionalities, and testing every function. The project will not be accepted if the freelancer implements only the examples listed in this job description. The project will also not be accepted if all decisions are implemented only as fixed Step Functions branches without meaningful AWS AgentCore observation, interpretation, decision-making, action selection, verification, and escalation. The AWS AgentCore agent must demonstrate this complete loop: * Observe * Understand * Decide * Act * Verify * Retry or escalate ## Mandatory Educational Documentation This is an educational project. Delivering working source code alone is not sufficient. The freelancer must teach me how the complete AWS AgentCore solution is designed, developed, deployed, tested, debugged, and operated. ### Step-by-Step DOCX Document The freelancer must provide a Microsoft Word DOCX file containing complete step-by-step instructions. The document must include a screenshot for every important step. It must cover: * Windows laptop prerequisites * Required software installation * AWS CLI installation and configuration * Python environment setup * AWS account preparation * Amazon S3 bucket creation * AWS IAM roles and policies * AWS Lambda development * AWS Step Functions development * Amazon CloudWatch configuration * Amazon Bedrock AgentCore configuration * AgentCore agent development * AgentCore tool development * Automatic trigger configuration * Deployment from a Windows laptop * Running the complete solution * Monitoring the agent * Reading logs * Simulating failures * Running all tests * Understanding agent decisions * Debugging problems * Verifying copied files * Reviewing reports and alarms * Human approval testing * Removing all created AWS resources Each step must include: * Purpose of the step * Exact AWS Console navigation * Exact Windows command * Source code or configuration used * Screenshot showing what should appear * Expected result * Common errors * Troubleshooting instructions Screenshots must not contain visible passwords, secret keys, access keys, tokens, or other confidential information. The DOCX file must be detailed enough for me to reproduce the complete project in my own AWS account from my Windows laptop. ### End-to-End Development Video The freelancer must provide a complete end-to-end instructional video. The video must teach the complete development process, not only demonstrate the finished application. The video must show: * Creating the project from the beginning * Explaining the architecture * Creating the project folders and source files * Developing the AgentCore agent * Developing the AgentCore tools * Developing the Lambda functions * Developing the Step Functions workflow * Creating IAM permissions * Creating S3 buckets * Configuring CloudWatch Logs * Configuring automatic execution * Deploying the solution * Running the first successful test * Creating failure simulations * Running every agent functionality test * Reading AWS results and CloudWatch logs * Explaining how the agent interprets each situation * Showing why the agent retries, stops, skips, waits, or asks for approval * Debugging failed tests * Correcting code or configuration * Re-running tests after corrections * Verifying checksums and file sizes * Reviewing reports and alarms * Testing human-in-the-loop approval * Cleaning up all AWS resources The freelancer must explain the work while performing it. The video must especially focus on: * AgentCore development * Agent tool development * Agent decision logic * Integration with Lambda and Step Functions * Testing every developed agent functionality * Failure simulation * Log interpretation * Debugging * Verification * Human-in-the-loop behavior A short video showing only the final successful execution will not be accepted. ### Documentation Acceptance Criteria The project will be accepted only when: 1. A complete Microsoft Word DOCX file is delivered. 2. The DOCX file contains step-by-step instructions for the entire project. 3. Every major development, deployment, configuration, and testing step includes a screenshot. 4. The DOCX file explains how to reproduce the project from a Windows laptop. 5. A complete end-to-end development video is delivered. 6. The video shows how the project is developed from the beginning. 7. The video demonstrates deployment to AWS. 8. The video demonstrates all developed agent functionalities. 9. The video demonstrates the complete test plan. 10. The video shows failure simulation, log review, agent decisions, recovery, and verification. 11. The video shows debugging of at least one intentionally failed scenario. 12. The video explains which operations are performed by deterministic AWS code and which decisions are made by AWS AgentCore. 13. The DOCX file and video include all additional functionalities and tests designed by the freelancer, not only the example scenarios supplied by me. 14. I can independently recreate, deploy, run, test, and troubleshoot the project in my own AWS account by following the DOCX file and video. 15. No additional paid support is required to reproduce the project.