whatbroke: CLI that captures context on test/dev server crashes for LLM debugging
whatbroke is a free open-source CLI that wraps any command (e.g., npm test) and, on a crash, captures a bundle of context: the error with parsed stack trace, the diff since the last passing run (recorded as a 'green' commit), and a ranked guess at the responsible file with reasons. This context is designed to be fed to an LLM (like Cursor) to enable accurate debugging, solving the problem of LLMs editing the wrong file when given only a stack trace.
Entities
Related
Git-aware AI debugger that checks out old commits to fix production bugs
A tool that makes AI coding assistants (like Cursor or Claude Code) automatically checkout the git commit corresponding to a production error before debugging, preventing the agent from analyzing current code that has shifted. It solves the problem of AI agents hallucinating fixes because they look at the present state of files while the bug existed in a past commit.
ContextOps: open-source tool to audit and optimize LLM prompt context
ContextOps is an open-source tool that analyzes LLM prompts to detect token waste such as duplicated retrieval chunks, bloated system prompts, oversized conversation history, and repeated tool outputs. It helps developers reduce costs and improve model consistency by auditing what goes into the prompt before inference.
Local CLI tool to analyze Claude AI token usage and costs
A Go-based CLI that parses local Claude AI logs to show token usage and cost breakdown by project, session, and model. It helps users understand where their API quota is going and how much they would have spent without caching.
temporal-debug-skill project enables AI agents to time-travel debug via git worktree
A new open-source project called temporal-debug-skill gives AI coding agents the ability to debug code at any point in git history. The skill parses fuzzy time references into commit SHAs, creates isolated read-only worktrees, and auto-cleans up, working with any agent like Claude Code or Cursor.
ddiff: LLM-generated diff format for isolating feature sets in code
ddiff is a format for isolating feature sets in code using LLM-generated diffs. It works by prompting an LLM to produce a diff of analysis of intents and code changes related to specific features, which can then be used by another LLM to implement the feature natively. The creator provides a live chat to Telegram group and a markdown WYSIWYG editor with rich uploads.