| name | suggest-fixes |
| description | Generate actionable suggestions for confirmed issues and low-grade evaluations |
| allowed-tools | Read, Bash, Glob, Grep |
| user-invocable | true |
Suggest Fixes
Generates actionable narrative suggestions for confirmed issues and evaluations that scored below A-. Runs as a separate step after investigation and evaluation, so the agent has full context.
Usage
/suggest-fixes
Prerequisites
- Issues should be investigated (
katz issue list --state confirmed returns results).
- Evaluations should be completed (
katz eval results returns results).
Workflow
1. Load confirmed issues
katz issue list --state confirmed
For each confirmed issue:
- Read the full issue with
katz issue show <id>
- Read the manuscript context around the issue's line range
- Write a specific, actionable suggestion for how to fix the problem
- Record with
katz issue suggest --id <id> --text "..."
2. Load low-grade evaluations
katz eval results
Filter for evaluations with grades below A- (B+, B, B-, C+, C, C-, D, F). For each:
- Read the criterion with
katz eval show <name>
- Read the evaluation response and grade
- Read the relevant manuscript section
- Write a specific suggestion for how to improve the grade
- Re-record with
katz eval respond --name <name> --text "..." --grade "..." --suggestion "..."
(preserve the existing text and grade, add the suggestion)
3. Guidelines for writing suggestions
- Be specific. "Improve the abstract" is useless. "Add a sentence after line 8 stating that no prior work has examined LLM responses for life preference revelation" is useful.
- Be actionable. The author should be able to act on the suggestion without further interpretation. Name the section, the line, the specific text.
- Be realistic. Don't suggest redoing the entire study. Suggest changes to the manuscript that address the identified concern.
- Propose language when possible. If the fix is a wording change, suggest the new wording. "Consider replacing 'potentially less prone to biases' with 'less noisy, though whether they are also less biased remains an open question.'"
- Prioritize. Not every confirmed issue needs an equally detailed suggestion. Focus effort on the issues that matter most for the paper's contribution.
4. Report
After generating suggestions, regenerate the HTML report to see them:
python <katz-skills-path>/find-issues/scripts/generate_review_report.py
Suggestions appear with a green left border and lightbulb icon in both issue and eval cards.
