Shell Script Reveals Which Claude Code Skills Are Draining Your Token Budget

·3 views

A developer has built a 106-line shell script called usage-breakdown.sh to identify which Claude Code Skills, Agents, and MCP servers are consuming the most resources during coding sessions. The tool addresses a gap in Claude Code's built-in /usage command, which only shows per-model cost totals without breaking down usage by individual tool or skill. The script parses transcript.jsonl files stored under ~/.claude/projects/ by counting tool_use events per skill, agent, and MCP server using Python's Counter. While it tracks call counts rather than exact token volumes, the author argues that call frequency serves as a reliable proxy for token consumption. The script also supports time-windowed analysis using file modification timestamps, allowing users to isolate recent sessions instead of mixing in older log data.

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)

Why 'I'll Start Later' Is a Trap: The Future Feels Easier Than It Is

A widely shared opinion piece on DEV Community argues that people consistently overestimate how much free time they will have in the future. The author explains that this happens because we experience the full weight of present responsibilities while imagining future life only through an idealized lens, free of stress and fatigue. In reality, responsibilities tend to accumulate over time rather than diminish, as school gives way to college, then work, and then broader life demands. The piece warns that delaying skill-building, such as learning to code, on the assumption that future circumstances will be more favorable is a flawed and potentially costly bet. The author urges readers to begin imperfectly today rather than wait for an ideal future moment that may never arrive.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer builds privacy-first Chrome extension that grades stocks with fixed rules

A developer has released StockAgent, a Chrome extension that scores stocks on a 0–5 scale using five deterministic financial metrics, including debt-to-equity ratio, RSI, and return on equity, with no AI involved in grading. News headlines from Yahoo Finance RSS feeds are also scanned for risk keywords, applying severity-based penalties to the final score. The tool offers optional email digests but is designed so that sensitive data — such as share counts, buy prices, and API keys — never leaves the user's device. To enforce this, the extension reconstructs outbound payloads from scratch using only pre-approved fields, rather than stripping private data from existing objects. A runtime validation function further blocks transmission if any forbidden field is detected, acting as a safeguard against accidental data leaks during development.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Healthcare AI Needs Forward-Deployed Engineers to Bridge Clinics and Code

Despite growing capability in building healthcare AI models, the medical field struggles to deploy them safely in real clinical settings, according to neurologist and software engineer Alireza Minagar. He argues that 'medical forward-deployed AI engineers' are needed to act as a bridge between technical teams and clinical staff, translating healthcare needs into workable system requirements. These specialists would oversee the full pipeline — from data ingestion to clinician decision-making — identifying failure points that standard technical monitoring cannot detect. Minagar also advocates for gradual deployment strategies, including silent model evaluation and mandatory kill switches to suspend AI systems when necessary. Without such human interfaces, he warns, even the most accurate AI models risk becoming clinically useless or silently unreliable once live.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

systemd vs BusyBox Init: How to Choose the Right Init System for Embedded Linux

Choosing between systemd and BusyBox init is one of the earliest and hardest-to-reverse decisions in embedded Linux product development, as it shapes service dependencies, failure detection, and library requirements for the product's entire lifespan. BusyBox init is generally recommended for devices with a fixed, small set of services, offering a minimal footprint and simple configuration, while systemd suits devices needing supervised, interdependent services with features like watchdog support and resource limits. Opting for systemd comes at the cost of a significantly larger image, a minimum kernel version of 5.10, and a glibc toolchain dependency. Developers should also note that init selection and device management are separate decisions — udev rules can be used without systemd in both Yocto and Buildroot build systems. A key BusyBox init limitation is that it lacks restart back-off or start limits, meaning a failing process will loop indefinitely, paced only by a one-second sleep in the init loop.

0 comments Read more at DEV Community