Conversation
Co-authored-by: Cursor <cursoragent@cursor.com>
- Add FAILURE RECOVERY AND COMMAND ECONOMY section to system prompt - Add FINAL ANSWER template section for cleaner solve completion - Add language-specific rules for Python, JS/TS, Shell/SQL - Enhance verification, error message, API, and edge-case guidance - Improve hail mary prompt to stress cost of empty patches Co-authored-by: Cursor <cursoragent@cursor.com>
Mechanical code changes (not prompt-only): - Parallelize git grep in _symbol_grep_hits and _broad_grep_fallback using ThreadPoolExecutor (saves 10-30s wall clock per attempt) - Fix false-negative test detection: use regex for non-zero fail counts instead of substring matching; add Go/Rust/zero-count good markers - Weighted per-file context budgets: top 3 ranked files get 2x budget for better edit-target coverage (SWE-Edit insight) - Raise stem-length floor from 3 to 5 in _rank_context_files to eliminate false positive matches on short words (SweRank insight) - Add out_of_time() check inside command batch loop to prevent wasting time on commands after budget is exhausted - Word-boundary matching for short tokens (<5 chars) in integration partner scoring to eliminate substring false positives Prompt additions (paired with mechanical changes): - FAILURE RECOVERY AND COMMAND ECONOMY section - FINAL ANSWER template - Python/JS/TS/Shell/SQL language-specific rules - Enhanced verification, error message, API, and edge-case guidance Co-authored-by: Cursor <cursoragent@cursor.com>
- Add path-term document-frequency downweighting in file ranking - Add symbol-centered preload excerpts for top symbol-hit files - Return symbol_hits from ranking for preload; extend docstring Co-authored-by: Cursor <cursoragent@cursor.com>
…prompts, wall-clock logging Co-authored-by: Cursor <cursoragent@cursor.com>
…rification Co-authored-by: Cursor <cursoragent@cursor.com>
Keep harness improvements: identifier stopwords/hook filter and self-check type cues. Co-authored-by: Cursor <cursoragent@cursor.com>
OpenRouter PR JudgeVerdict: PASS
SummaryThis PR modifies the preloaded context ranking and reading path in agent.py: it adds IDF-style document-frequency downweighting for path-term matches, introduces excerpt-centering around the first issue-symbol match in the top files, allocates per-file budgets non-uniformly across ranked files, parallelizes git grep calls in _broad_grep_fallback and _symbol_grep_hits via ThreadPoolExecutor, replaces the prior _issue_identifier_path_boost with the new IDF logic, adds a mid-batch wall-clock check, tightens _looks_like_successful_test_output with a non-zero failure regex and additional good/bad markers, and expands the system / self-check / hail-mary prompts with anti-partial-ship and wiring-audit guidance. Static Checks
Judge Reasons
Risks
Required Changes
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
5HDqwhrHkPH2eh4atZjseCLpWM58ZdJ1DeLbQpSMzYCq8TMa