fix(dedup): clarify tool sequencing in prompts to fix regression

Explicitly instruct the model to call the domain tool first, then echo the output, to avoid it skipping the tool call.
This commit is contained in:
cocosheng-g
2026-02-03 21:24:55 -05:00
parent 36ce66933e
commit b4762750e2
2 changed files with 3 additions and 3 deletions

View File

@@ -121,9 +121,9 @@ jobs:
## Steps
1. **Find Potential Duplicates:**
- The repository is ${{ github.repository }} and the issue number is ${{ github.event.issue.number }}.
- Use the `duplicates` tool with the `repo` and `issue_number` to find potential duplicates for the current issue. Do not use the `threshold` parameter.
- First, call the `duplicates` tool with the `repo` and `issue_number` to find potential duplicates. Do not use the `threshold` parameter.
- Then, AFTER the tool completes, use the `echo` command to print the tool's JSON output to the logs.
- If no duplicates are found, you are done.
- Use the `echo` command to print the JSON output from the `duplicates` tool to the logs.
2. **Refine Duplicates List (if necessary):**
- If the `duplicates` tool returns between 1 and 14 results, you must refine the list.
- For each potential duplicate issue, run `gh issue view <issue-number> --json title,body,comments` to fetch its content.

View File

@@ -106,7 +106,7 @@ jobs:
1. **Extract Repository Information:** The repository is ${{ github.repository }}.
2. **Refresh Embeddings:** Call the `refresh` tool with the correct `repo`. Do not use the `force` parameter.
3. **Log Output:** Use the `echo` command to print the JSON output from the `refresh` tool to the logs.
3. **Log Output:** After the `refresh` tool completes, use the `echo` command to print the tool's JSON output to the logs.
## Guidelines