fix: ensure agent responses are replies to triggering comments

## Summary
Updated the prompt and code to ensure that when the action is triggered by a user comment, the agent's responses are actual **replies** to that specific comment rather than standalone comments.

## Changes
- Updated `src/utils/prompt.ts` with explicit instructions for the agent to reply to comments using the triggering comment ID
- Added `commentId` to `AuggieParams` type in `src/utils/index.ts`
- Pass `commentId` through `src/index.ts` to `runAuggie`
- Include triggering comment ID in the prompt context in `src/auggie.ts`

## Why
Previously, the agent would sometimes create regular comments instead of threaded replies, making conversations harder to follow. This change ensures consistent behavior where responses appear as replies to the original triggering comment.

Author: tulga-bytes

src/auggie.ts

@@ -14,6 +14,8 @@ export type RunAuggieOptions = {
   context?: GithubPullRequest;
   /** The body of the comment that triggered this action, if triggered by a comment */
   commentBody?: string;
+  /** The ID of the comment that triggered this action, if triggered by a comment */
+  commentId?: number;
 };
 
 /**
@@ -27,6 +29,7 @@ export async function runAuggie(options: RunAuggieOptions): Promise<string> {
     workspaceRoot,
     context,
     commentBody,
+    commentId,
   } = options;
 
   const workspace = workspaceRoot || process.cwd();
@@ -70,6 +73,12 @@ export async function runAuggie(options: RunAuggieOptions): Promise<string> {
       core.info("📨 User comment included in prompt");
     }
 
+    // Add triggering comment ID if available
+    if (commentId) {
+      fullPrompt = `${fullPrompt}\n\n## Triggering Comment ID:\n${commentId}`;
+      core.info(`📍 Triggering comment ID: ${commentId}`);
+    }
+
     // Add user prompt
     fullPrompt = `${fullPrompt}\n\n${userPrompt}`;
 

src/index.ts

@@ -19,6 +19,7 @@ async function main(): Promise<void> {
 		augmentApiUrl,
 		workspaceRoot,
 		commentBody,
+		commentId,
 	} = getAuggieParams();
 	const { owner, repo } = parseRepository();
 	const githubToken = process.env.GITHUB_TOKEN;
@@ -34,6 +35,7 @@ async function main(): Promise<void> {
 		apiUrl: augmentApiUrl,
 		workspaceRoot: workspaceRoot || undefined,
 		commentBody,
+		commentId,
 	});
 
 	core.info("✅ Auggie agent completed successfully");

src/utils/index.ts

@@ -169,6 +169,7 @@ type AuggieParams = {
   augmentApiUrl: string;
   workspaceRoot: string | undefined;
   commentBody: string | undefined;
+  commentId: number | undefined;
 };
 
 /**
@@ -189,12 +190,14 @@ export function getAuggieParams(): AuggieParams {
   const augmentApiUrl = getInput("augment_api_url", true);
   const workspaceRoot = getInput("workspace_root");
   const commentBody = getCommentBodyFromEvent();
+  const commentId = getCommentIdFromEvent();
   return {
     eventName,
     prompt,
     augmentApiKey,
     augmentApiUrl,
     workspaceRoot,
     commentBody,
+    commentId,
   }
 }

src/utils/prompt.ts

@@ -6,11 +6,21 @@ Important: How you handle changes depends on the context:
 - If you are working on a Pull Request (PR): You are allowed to commit directly to that PR's branch.
 - If you are working on a GitHub Issue: You must create a new branch and open a Pull Request with your changes.
 
+CRITICAL - Replying to Comments:
+When this action is triggered by a user comment, you MUST reply to that specific comment (not create a new standalone comment).
+- A "Triggering Comment ID" will be provided in the prompt context.
+- Use this comment ID to create a REPLY to that comment using the GitHub API.
+- For issue comments: Use POST /repos/{owner}/{repo}/issues/comments/{comment_id}/replies or reference the comment in your reply.
+- For PR review comments: Use POST /repos/{owner}/{repo}/pulls/{pull_number}/comments with the \`in_reply_to\` parameter set to the triggering comment ID.
+- This ensures your response appears as a threaded reply to the user's original comment, making the conversation easy to follow.
+- If no triggering comment ID is provided, you may create a new comment.
+
 Your workflow:
-1. Add a comment to the PR letting the user know you're starting to work on it.
+1. Reply to the triggering comment to let the user know you're starting to work on it.
+   - IMPORTANT: Use the Triggering Comment ID to reply directly to that comment (see above).
 2. Read the PR context and understand the request.
 3. Create a Todo List:
-   - Update the original comment on the PR with a todo list of what you're going to do. (Use the github-api tool to update the comment. Replace the initial message with the todo list.)
+   - Update your reply comment with a todo list of what you're going to do. (Use the github-api tool to update your reply comment.)
    - Use your GitHub comment to maintain a detailed task list based on the request.
    - Format todos as a checklist (- [ ] for incomplete, - [x] for complete).
    - Update the comment with each task completion.
@@ -29,7 +39,7 @@ Your workflow:
      - Files modified
      - Links to PRs/commits
 5. Final Comment Update (IMPORTANT):
-   - When you are completely done with all tasks, update the original comment ONE FINAL TIME.
+   - When you are completely done with all tasks, update your reply comment ONE FINAL TIME.
    - REMOVE the entire task list/checklist from the comment.
    - Replace the comment body with ONLY a concise, minimal final summary.
    - Keep it brief: just state what was attempted and what was done.