Developer builds terminal snapshot tool to restore Claude Code sessions after crashes

·4 views

A developer has released a tool called claude-sessions that snapshots the full state of an iTerm terminal every five minutes using a launchd daemon and AppleScript. The snapshot records open tabs, working directories, and any active Claude Code sessions, storing up to 100 distinct states as a history file. After a machine crash, users can run a single command to restore all tabs to their previous directories and resume Claude sessions automatically. The developer deliberately avoided auto-restore, making users the explicit trigger to distinguish crashes from intentional quits. The project, available on GitHub, also includes a cross-project session picker that lets users resume any Claude Code session from any repository in a single interface.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How Claude Code Uses Rules, Skills, and Agents to Enforce SQL Conventions

AI coding agent Claude Code can generate stored procedures, migration scripts, and complex SQL reports rapidly, but unreviewed generated code risks drifting from established team conventions. A practical framework addresses this through three mechanisms: rules files that automatically apply coding standards at generation time, slash commands that encapsulate repeatable tasks, and agents that orchestrate multi-step data workflows. Rules files stored in a repository are passed to Claude Code on every request, effectively replacing a style guide that relies on human memory with one that is machine-enforced by default. The approach is designed specifically for SQL and ETL work, where recurring patterns such as naming conventions and procedure layouts are common enough for reliable AI reproduction. The framework emphasizes that enforced conventions and human understanding of the underlying business question must work together for AI-assisted SQL development to be worthwhile.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI Agent Projects Often Fail Due to Poor Data Prep, Not Model Limitations

A growing number of AI agent projects run over schedule and budget not because of model failures, but because of unresolved data problems that teams fail to plan for upfront. Engineers typically scope costs around tokens, vector stores, and orchestration, while overlooking the significant effort required to clean, deduplicate, and govern the underlying data. There are two distinct data challenges: knowledge data, such as documents and FAQs, which breaks down due to format inconsistency and terminology drift, and operational data, such as customer records, which fails due to identity resolution and access authority gaps. Unresolved data issues can cause agents to return confidently wrong answers, as illustrated by a demo where the same pipeline query returns wildly different totals depending on how many flawed records are retrieved. Experts recommend teams honestly assess whether a competent human could answer the agent's intended questions from existing data before committing to any AI agent build.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Accrual vs Cash Accounting: Why the Difference Matters for Cross-Border API Builds

Software developers building accounting integrations must understand the distinction between cash-based and accrual accounting, as the method a business uses fundamentally changes what financial data means. Cash-based accounting records revenue and expenses only when money physically moves, making it simple but potentially misleading about a company's true financial performance across periods. Accrual accounting recognizes revenue when earned and expenses when incurred, offering a more accurate long-term picture and forming the basis of both GAAP and IFRS reporting standards. The rules governing which businesses must use accrual accounting differ significantly across the US, UK, and Europe, with thresholds, defaults, and mandates varying by jurisdiction and company size. Developers pulling financial data through APIs without knowing the accounting basis risk building flawed risk models, forecasts, or dashboards based on misunderstood figures.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Dev Team Builds 2ms Discord Moderation Pipeline in .NET Without Generative AI

A development team managing a Discord server with over 50,000 concurrent users built a custom real-time content moderation pipeline in C# .NET after finding that standard bots and generative AI models introduced unacceptable latency, hallucinations, and compliance risks. The system abandons third-party large language models in favor of a zero-allocation, constant-time architecture that processes incoming messages using stack-allocated memory buffers instead of heap-based string objects. To keep AI inference fast, the team engineered a compression heuristic that reduces any message to a 64-character window by extracting the first and last 32 characters, enabling consistent vector embedding generation regardless of message length. A local dual MiniLM-L6-v2 model pipeline then evaluates semantic intent, with toxicity classifications requiring a confidence score between 0.95 and 0.98 to reduce false positives. The result is a moderation pipeline that achieves sub-12ms execution latency with deterministic, auditable behavior and no reliance on external AI services.

0 comments Read more at DEV Community