[WIP] Fix issue download script to use GitHub MCP tools

.github/workflows/chroma-issue-indexer.md

@@ -26,37 +26,45 @@ This workflow indexes issues from the repository into a Chroma vector database f
 
 Index the 100 most recent issues from the repository into the Chroma vector database:
 
-1. **Fetch Issues**:
-   - Use the GitHub MCP server tools to list the most recent 100 issues
-   - Include both open and closed issues
-   - Get issue number, title, body, state, created date, and author
-
-2. **Create/Update Chroma Collection**:
-   - Create a collection named "issues" if it doesn't exist (use `chroma_create_collection`)
-   - Use an appropriate embedding function for semantic search
-
-3. **Index Issues**:
-   - For each issue, add it to the Chroma collection (use `chroma_add_documents`)
-   - Use ID format: `issue-{issue_number}`
-   - Document content should be: `{title}\n\n{body}` (title and body combined)
-   - Include metadata:
-     - `number`: Issue number
+1. **Create Chroma Collection First**:
+   - IMPORTANT: Check if the "issues" collection exists using `chroma_list_collections`
+   - If it doesn't exist, create it using `chroma_create_collection` with:
+     - Collection name: "issues"
+     - Use default embedding function (omit embedding_function_name parameter)
+
+2. **Fetch Issues Using GitHub MCP Tools** (NOT Python scripts):
+   - Use the `list_issues` tool from GitHub MCP server to fetch issues
+   - Fetch issues in batches of 5 at a time using the `perPage: 5` parameter
+   - Start with page 1, then page 2, page 3, etc. until you have 100 issues total
+   - Include both open and closed issues (omit state parameter to get both)
+   - Order by created date descending to get most recent first: `orderBy: "CREATED_AT"`, `direction: "DESC"`
+   - For each issue, extract: number, title, body, state, createdAt, author.login, url
+
+3. **Index Issues in Batches**:
+   - Process each batch of 5 issues immediately after fetching
+   - For each batch, use `chroma_add_documents` to add all 5 issues at once
+   - Use ID format: `issue-{issue_number}` (e.g., "issue-123")
+   - Document content: `{title}\n\n{body}` (combine title and body)
+   - If body is empty/null, use just the title as content
+   - Include metadata for each issue:
+     - `number`: Issue number (as string)
      - `title`: Issue title
      - `state`: Issue state (OPEN or CLOSED)
      - `author`: Issue author username
-     - `created_at`: Issue creation date
+     - `created_at`: Issue creation date (ISO 8601 format)
      - `url`: Issue URL
 
 4. **Report Progress**:
-   - Log how many issues were indexed
-   - Note any issues that couldn't be indexed (e.g., empty body)
-   - Report the total number of issues in the collection
+   - After processing all batches, use `chroma_get_collection_count` to get total issue count
+   - Report how many issues were successfully indexed
+   - Note any issues that couldn't be indexed (e.g., API errors)
 
 ## Important Notes
 
-- Process issues in batches if needed to avoid rate limits
-- Skip issues that have already been indexed (check if ID exists)
-- For issues with empty bodies, use just the title as content
+- **MUST use GitHub MCP tools** (`list_issues` tool), NOT Python scripts or `gh` CLI
+- **MUST create collection first** before attempting to add documents
+- Process exactly 5 issues per batch using `perPage: 5` and incrementing page number
+- Skip duplicate issues (Chroma will update if ID exists)
 - The collection persists in `/tmp/gh-aw/cache-memory-chroma/` across runs
 - This helps other workflows search for similar issues using semantic search