Claude Code merges custom slash commands into skills, adding new capabilities

·1 views

Anthropic's Claude Code has merged custom slash commands into a unified 'skills' system, a change verified against official documentation as of late July 2026. Previously, users could create simple slash commands by dropping markdown files into a .claude/commands/ directory; those files continue to work, but the underlying model has been consolidated with the skills framework. The merge introduces new features such as supporting file directories, frontmatter-controlled invocation permissions, and the ability for Claude to automatically trigger a skill when a conversation matches its description. Key behavioral details include shell-quoted indexed argument expansion via $ARGUMENTS, one-time tool grants that expire after a single turn, and shell command substitution using the !` ` syntax to inject live data like git diffs before Claude processes a prompt. When a command and a skill share the same name, the skill takes precedence, and developers are advised to reserve the commands directory for personal shortcuts while using skills for anything intended to grow or be shared.

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)

Two-stage image pipeline offers reliable character consistency without LoRA training

Developers building AI apps that generate the same character across multiple scenes often struggle with face drift when using seeds or prompt-only methods. A two-stage pipeline addresses this by first generating one canonical base image, then using an image-edit model with that base as a reference for every subsequent scene. Unlike seeds, which only reproduce identical inputs, or LoRA fine-tuning, which requires curating dozens of images and running training jobs per character, this approach conditions each new generation on the actual reference pixels. The edit model is instructed only on what should change — pose, setting, lighting — while identity is preserved through the input image itself. This method is described as scalable for production apps where users create characters on demand, and works for non-human subjects such as animals, robots, or animated characters as well.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds 28-rule prompt injection firewall to protect AI agents from MCP threats

Edison Flores of AliceLabs LLC has released L1.9, a prompt injection defense layer designed to scan AI agent inputs before potentially malicious content enters a large language model's context window. The tool targets a known vulnerability in MCP (Model Context Protocol) servers, where tool descriptions can carry hidden instructions to override agent behavior or exfiltrate sensitive data. L1.9 applies 28 detection rules across skill names, descriptions, system prompts, and capability schemas, flagging threats at critical, high, or medium severity levels and quarantining dangerous skills automatically. Each detected finding is mapped to a MITRE ATT&CK technique ID and includes a snippet of the offending text for transparency. Flores claims L1.9 is part of a broader 10-layer security stack, which he says is unmatched by most existing MCP directories that offer no such protections.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Accurately Convert Bank Statement PDFs to CSV: Methods Compared

Converting bank statement PDFs to CSV is error-prone due to structural issues like merged cells, multi-line transactions, and inconsistent debit/credit column formats that vary by bank. A basic manual copy-paste method works but is slow and unreliable, taking 20–40 minutes per statement with high risk of errors. Developers can use Python's pdfplumber library for a scriptable solution, though it fails on scanned or image-based PDFs and requires bank-specific customization. AI-powered converters offer the most reliable results across different banks and formats, including scanned documents, by interpreting document semantics rather than detecting table boundaries. Regardless of method used, output quality should be verified by checking row counts, amount reconciliation, sign consistency, and the handling of multi-line descriptions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Sovereign Package Released Free on GitHub to Deploy Lemmings Across Cloud Regions

A developer has released Sovereign, a free open-source package on GitHub designed to deploy Lemmings load-testing tools across up to three cloud regions. The tool runs dry-runs to estimate costs and consolidates results into a single aggregated report as load test instances spin up and down. Sovereign is aimed at organizations using AI to build applications who want to stress-test their infrastructure before committing to large marketing or development budgets. The package is fully self-serve, requiring no involvement from the author, with guidance provided through detailed README files and code comments. The developer notes it is available now at no cost, with optional paid consulting support for those who need it.

0 comments Read more at DEV Community