Claude Code Plugins Expose Context Costs

// 1h agoPRODUCT UPDATE

Claude Code Plugins Expose Context Costs

Claude Code now documents how plugins consume always-on and per-invocation context, even when their skills or agents never run. Developers can inspect usage, identify stale plugins, and disable or uninstall them to reclaim budget and reduce attack surface.

// ANALYSIS

Plugin ecosystems are becoming context-budget ecosystems: every convenience carries a persistent tax, so visibility and cleanup are essential.

–Skill, agent, and command descriptions enter context on every turn; full component content loads only when invoked.
–`claude plugin details` separates always-on tokens from per-invocation costs, while official marketplace listings expose estimates before installation.
–`/plugin`, `/skill-doctor`, `/doctor`, and `/usage` help locate unused or expensive extensions; plugins unused for 14 days across 10 sessions are flagged.
–MCP servers and hooks can run with the user’s permissions, making plugin review a security task as well as a cost-optimization task.
–For teams, OpenTelemetry and Anthropic’s Analytics API extend plugin usage measurement across organizational deployments.

// TAGS

claude-codecoding-agentai-codingagentcontext-engineeringmcpdevtoolsecurity

DISCOVERED

1h ago

2026-10-10

PUBLISHED

1h ago

2026-10-10

RELEVANCE

9/ 10

AUTHOR

DIY Smart Code

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

TUTORIAL53m ago

Grok Bot showcases 140% pipeline growth

An upcoming Oct. 13 webinar will feature Above the Line’s claim that a team of Grok Bots grew its sales pipeline by 140% in one month and helped book 11 events through prospect research and outreach drafts.

INFRA1h ago

National Compute Opens B300 Access

National Compute is offering eligible .edu, .gov, and .mil users access to whole eight-GPU NVIDIA B300 or AMD MI355X nodes, with $100 in starter credits and no card required at signup. Researchers connect over SSH for open-weight inference, LoRA fine-tuning, image-model training, and other experiments.

MODEL2h ago

Solar Mini 4 Makes Cheap Agents Convincing

Upstage’s Solar Mini 4 is a 35B-parameter mixture-of-experts model with roughly 3B active parameters, a 512K context window, tool calling, and structured outputs. It is temporarily free in Hermes Agent, making it attractive for research, extraction, and repetitive workflows. [Upstage](https://www.upstage.ai/blog/en/solar-mini-4)