### Background To enrich Workflow use cases and to leverage to infrastructure we created we want to speed run adding more tools to workflow ### Tools and Priorities Table contains the tools, and use cases they will enable. Priority marks the order in which we should go about adding them where 1 represents the most important ones <table> <tr> <td> **Tool name**  </td> <td> **Use cases it enables**  </td> <td> **Stage**  </td> <td> **Fail cases and responses**  </td> <td> **Priority**  </td> </tr> <tr> <td>Read file </td> <td> * WF can get awareness on a CB so it can operate on it  </td> <td>Execution </td> <td> * File might not exist or agent might not have access to it. </td> <td>Already exists </td> </tr> <tr> <td>Write file </td> <td> * WF can make file changes. WF can create and generate code for the user   </td> <td>Execution </td> <td> </td> <td>Already exists </td> </tr> <tr> <td>Run command </td> <td> * WF can work at the OS level * Can run tests to make sure they pass  * Get better CB understanding for planing   </td> <td>Planning, Execution   </td> <td> * Command might ask for user input on std in blocking tool execution (related to https://gitlab.com/gitlab-org/duo-workflow/duo-workflow-service/-/issues/44#note_2045395573) </td> <td>Already exists </td> </tr> <tr> <td>Search API</td> <td> * WF can access SCM, Epics, Issue, MR’s  * Get this issue and solve it  * Find duplicate code and refactor it   </td> <td>Planing </td> <td> * Agent get confused because when presented with multiple, or irrelevant search results * Time out is a rare risk but possible   </td> <td>Already Exists </td> </tr> <tr> <td>Create, Read, Update  MR </td> <td> * WF can participate in a code review  * WF can author new MR’s from scratch  </td> <td>Planing, Execution </td> <td> * There is a risk of agent creating excessive and redundant entities </td> <td>Creating a MR has not yet been implemented. Reading and updating an existing MR exists</td> </tr> <tr> <td>Create,Read, Update Issue </td> <td> * WF can prepare a refactor plan and to create issue that will document the process  </td> <td>Planing, Execution </td> <td> * There is a risk of agent creating excessive and redundant entities </td> <td>Creating and reading issues tool exists. Updating issues tool is incoming shortly along with the ability to write comments in notes.</td> </tr> <tr> <td>Create, Read, Update Epic</td> <td> * WF can create an Epic and than create all the issues as child work items  </td> <td>Planing, Execution </td> <td> * There is a risk of agent creating excessive and redundant entities </td> <td>5</td> </tr> <tr> <td>Web browsing </td> <td> * WF can visit web pages to retrieve data from them  </td> <td>Planing, Execution </td> <td> * Timeout and slow responses might decrease WF performance * Irrelevant results might confuse agent * Risk of adversary interaction using fabricated inputs (see https://www.businessinsider.com/google-ai-glue-pizza-i-tried-it-2024-5?IR=T) </td> <td>6</td> </tr> <tr> <td>Notes API </td> <td> * Enables WF to insert comments into different model  </td> <td>Planing, Execution </td> <td> * There is a risk of agent creating excessive and redundant entities </td> <td>Inserting/reading notes from MR exists. Issues and epics incoming 3</td> </tr> <tr> <td>Consuming GitLab Public API</td> <td> * WF can interact with all of GitLab  * WF could optimize prioritize etc backlogs for planing  * Review all issues that are more than X old and decide if they are even relevant to the current CB  * Start to code all issues that are weighted X  </td> <td>Planing, Execution </td> <td> * To broad access to API endpoints that mutates state of GitLab can create risk of agent causing harm with hallucinated actions </td> <td>Long term, spike is required to flash out full scope </td> </tr> </table>