We're Adding a New Plugin to Claude Code: You Should...  🔗 **Source:** [We're Adding a New Plugin to Claude Code: You Should...](https://twitter.com/ClaudeDevs/status/2106118517447876618/?rw_tt_thread=True) > [!note]+ Document Notes > tags: [claude-code, claude-code-mods, developer-tools, agent-supervision, plugins] > author: ClaudeDevs > source: X (Twitter), @ClaudeDevs > reference: https://twitter.com/ClaudeDevs/status/2106118517447876618 > date created: 2026-10-04 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > <!-- deep-read:sections start --> ## 🏷️ Title **"We're adding a new plugin to Claude Code: You should Know."** — @ClaudeDevs, two-post thread on X, October 2, 2026. 73 words plus a demo video. Announces a built-in Claude Code plugin that "scans Claude's output for important information you might miss," enabled with `/plugin enable cc-plugin-you-should-know@builtin`. The second post says it's a **mod** that spins off a "sideagent" to observe output, and links the mods tutorial. **Disclosure.** This analysis is produced by a Claude model; the post is Anthropic's own product announcement. **Context fetched.** The linked post, "Getting started with Claude Code mods" (Addy Osmani, claude.dev, October 1, 2026, 11-minute read), was fetched for context. It defines a mod as "a small JavaScript or TypeScript file that runs inside your Claude Code session," shipped as a plugin whose `hooks/hooks.json` names a module exporting `register(on, options)`, with hooks that form a middleware chain (`next(e)`). Requires Claude Code 2.1.287 or later; mods are on by default. The demo video wasn't watched. ## 🧭 Table of Contents 1. [References (in-page)](#-references-in-page) 2. [Summary](#-summary) 3. [Main Content](#-main-content) → [Notes](#️-notes) 4. [References](#-references) 5. [Paraphrase](#️-paraphrase) 6. [Newsletter](#-newsletter) 7. [Action Items](#-action-items) 8. [Hypotheses](#-hypotheses) 9. [Methodology](#-methodology) 10. [Questions](#-questions) 11. [Future Research](#-future-research) 12. [Implications](#-implications) 13. [Learning Styles Assessment](#-learning-styles-assessment) 14. [Accessibility Assessment](#-accessibility-assessment) 15. [DEIB Assessment](#-deib-assessment) 16. [Logical Validity](#-logical-validity) 17. [DARVO Analysis](#️-darvo-analysis) 18. [Conclusion](#-conclusion) 19. [Refactor Appendix](#-refactor-appendix) ## 🔗 References (in-page) The markup was enumerated before any sourcing claim was made. One outbound link: 1. **https://claude.dev/blog/getting-started-with-claude-code-mods/** — first-party tutorial (Addy Osmani, Oct 1, 2026). Fetched; covers mod structure, the hook chain, a worked "Token Weather" example (~80 lines), and two larger examples ("Blast Radius," "Replay Theater"). Plus one embedded video (`amplify_video/2106118378331271168`), not watched. No link to the You Should Know plugin's source or docs. ## 📝 Summary - **What happened.** Anthropic added a built-in Claude Code plugin, "You should Know," that runs a side agent over Claude's output and surfaces important items a user might miss. - **Why it matters.** It's a first-party example of the new mods system and of a supervision pattern — a second agent watching the first — packaged as a one-line enable. - **Key stakeholders.** Claude Code users who skim long outputs; mod authors; teams building agent-supervision tools. - **Risks and tradeoffs.** A side agent costs tokens; what counts as "important" isn't described; a second model's summary can itself miss or misstate things. The post doesn't say what the plugin does with findings. - **Bottom line.** A built-in mod that watches Claude's output and flags what you'd otherwise skip; useful if you skim, unproven on accuracy. ## 📖 Main Content **What it does.** "Scans Claude's output for important information you might miss to help keep you in the loop." **How it's enabled.** `/plugin enable cc-plugin-you-should-know@builtin` — the `@builtin` suffix marks a plugin shipped with Claude Code rather than from a marketplace. **How it works.** "A mod that works by spinning off a sideagent to observe Claude's output." In the mods API (per the linked tutorial), hooks receive every session event and can observe (`await next(e)` then inspect), rewrite, or answer events. An observing hook on model output can pass that output to a secondary model call and draw UI with the result. **Mods, from the linked tutorial.** - A mod is a plugin with `.claude-plugin/plugin.json` and `hooks/hooks.json` naming a module. - The module exports `register(on, options)`; `on(event, matcher?, hook)` adds a hook. - Hook signature: `async ($, e, next) => …`, where ` exposes `ui, session, state, store, fs, process, clock, http, tool, command, model`. - Three behaviors: answer (return without calling `next`, e.g., `{ deny }`), observe, rewrite. - Claude Code writes the mod's type declarations into `.claude-plugin/types/` on load; those types are the authority for the installed version. **Terms.** - **Mod** — a Claude Code plugin with JS/TS hook modules that can observe, change, or draw UI. - **Side agent** — a secondary agent run alongside the main session, here used as an observer. **Unknowns.** What "important" means operationally; how findings are shown; token cost per turn; whether it can be configured. ## 🗒️ Notes - The name "You should Know" frames the output as a nudge, not a gate. - "information you might miss" — the problem is attention, not correctness. Long agent outputs bury caveats ("I couldn't run the tests," "I skipped X"). - "to help keep you in the loop" — human-in-the-loop framing. - `@builtin` — ships in the product; no marketplace install. - "spinning off a sideagent" — confirms a second model call, which means extra cost. - The thread uses the plugin as an advertisement for mods: the second post's job is the blog link. - The tutorial's "Four habits worth keeping" and "Sharing your mod" sections suggest Anthropic expects community mods. ## 📚 References - @ClaudeDevs thread (source) — https://twitter.com/ClaudeDevs/status/2106118517447876618 - Addy Osmani, "Getting started with Claude Code mods," claude.dev, Oct 1, 2026 — https://claude.dev/blog/getting-started-with-claude-code-mods/ - Claude Code documentation — https://docs.claude.com/en/docs/claude-code/overview - Claude Code plugins documentation — https://docs.claude.com/en/docs/claude-code/plugins ## 🗣️ Paraphrase - **Post 1.** Claude Code has a new built-in plugin called You should Know. It reads Claude's output and points out important things you might skip, so you stay informed. Turn it on with one slash command. (Demo video attached.) - **Post 2.** It's a mod: it starts a side agent that watches what Claude writes. The new blog post explains mods. ## 📰 Newsletter **Claude Code now has a built-in "did you catch that?" plugin** Anthropic shipped **You should Know**, a Claude Code plugin that runs a side agent over Claude's output and flags things you might skim past. Turn it on with `/plugin enable cc-plugin-you-should-know@builtin`. It's also a showcase for **mods** — small JS/TS hook files that can watch, rewrite, or add UI to a Claude Code session. Addy Osmani's [getting-started guide](https://claude.dev/blog/getting-started-with-claude-code-mods/) builds one from scratch in about 80 lines. WordPress developers using Claude Code for plugin work: the obvious mod to build is one that flags when an agent edits a file touching `$_POST`, `$wpdb->query`, or `echo` without escaping. The hook API makes "watch for this and tell me" a short file. ## ✅ Action Items #### DevRel - **Developer experience:** write a WordPress-specific mod example (e.g., flag unescaped output or missing nonce checks in agent edits) to show mods in a familiar domain. - **Education:** the "observer agent" pattern is teachable on its own; a short demo comparing output with and without You Should Know would show what it catches. #### OSPO - **Security Management:** mods run code in the session with access to `fs`, `process`, `http`; treat third-party mods like any plugin — review before install. - **Software Inventory:** track which mods are enabled across team machines. #### Community - **Contributors:** a community gallery of reviewed mods would reduce duplicated effort. #### WordPress - **Core:** Not applicable. - **Core Performance:** Not applicable. - **Core AI:** the observer-agent pattern parallels proposals for AI review assistants in the editor; worth noting as prior art. - **Core Program Team:** Not applicable. - **Mobile:** Not applicable. - **Playground:** Not applicable. - **Community:** Not applicable. - **Training:** Not applicable. - **Marketing:** Not applicable. - **Support:** Not applicable. - **Design:** Not applicable. - **Documentation:** Not applicable. - **Photos:** Not applicable. - **TV:** Not applicable. - **Polyglots:** Not applicable. - **Accessibility:** Not applicable. - **Meta:** Not applicable. - **CLI:** Not applicable. - **Hosting:** Not applicable. - **Tide:** Not applicable. - **Openverse:** Not applicable. - **Themes:** Not applicable. - **Plugins:** a mod that runs WordPress Coding Standards or Plugin Check on files an agent edits would catch issues before commit. #### The WPOCC - Not applicable to this source. #### FAIR - Not applicable to this source. ## 🧪 Hypotheses 1. **H1.** Users with You Should Know enabled will catch more agent-reported caveats (skipped tests, unverified steps) than users without it. 2. **H2.** The plugin adds a measurable per-turn token cost, scaling with output length. 3. **H3.** The side agent's flags have a false-positive rate high enough that some users disable it within a week. ## 📊 Methodology - **H1:** Seed sessions with outputs containing known caveats; measure user recall with and without the plugin. - **H2:** Compare `/cost` or usage logs across matched sessions. - **H3:** Opt-in telemetry on enable/disable events, or a user survey. ## ❓ Questions - What criteria does the side agent use for "important"? - Which model does the side agent run on? - Can users configure what it watches for? - How does it show findings — a band, a toast, inline? - How does it interact with existing hooks and other mods? ## 🔮 Future Research - Effectiveness of observer agents in reducing human oversight failures. - Cost-benefit of supervision agents vs. better primary-agent reporting. ## 💡 Implications - **Practice.** Users who skim long outputs get a safety net. - **Policy.** Not applicable. - **Security.** The mods system's reach (`fs`, `process`, `http`) makes third-party mods a supply-chain surface. - **Education.** A simple, visible example of agent-on-agent supervision. - **Funding.** Not applicable. ## 🧠 Learning Styles Assessment - **Reading level estimate:** Flesch-Kincaid Grade ≈ 8. #### 👀 Visual learners - **Present:** demo video. - **Improve:** a screenshot of a flagged item. #### 🔊 Aural learners - **Present:** none. - **Improve:** Not applicable. #### 🗣️ Verbal learners - **Present:** plain description. - **Improve:** define "mod" and "sideagent" in the post. #### 🧩 Kinesthetic learners - **Present:** a one-line enable command. - **Improve:** Not applicable. #### 🧠 Logical learners - **Present:** none. - **Improve:** criteria and cost figures. #### 🧑‍🤝‍🧑 Social learners - **Present:** replies on X. - **Improve:** Not applicable. #### 🧘‍♀️ Solitary learners - **Present:** a linked tutorial for self-paced learning. - **Improve:** Not applicable. #### 🌿 Naturalistic learners - **Present:** none. - **Improve:** Not applicable to this source. ## 📊 Accessibility Assessment #### Screen reader semantics - **Present:** text post; link text is a truncated URL ("claude.dev/blog/getting-s…"). - **Improve:** use the post title as link text. #### Keyboard navigation - **Present:** platform-dependent. - **Improve:** Not applicable. #### Contrast and readability - **Present:** platform default. - **Improve:** Not applicable. #### Text resizing and reflow - **Present:** platform reflow. - **Improve:** Not applicable. #### Media and captions - **Present:** demo video; captions unknown. - **Improve:** add captions or a text description of what the demo shows. #### Forms and interactions - **Present:** none. - **Improve:** Not applicable. #### Standards reference - **WCAG 2.2 AA 2.4.4 Link Purpose** — truncated URL as link text is weak; **1.2.2 Captions** — unverified for the video. ## 🌍 DEIB Assessment #### Cultural breadth - **Present:** Not applicable. - **Improve:** Not applicable. #### Inclusive language - **Present:** neutral. - **Improve:** Not applicable. #### Neurodiversity support - **Present:** the plugin itself is an attention aid — useful for users who lose details in long outputs. - **Improve:** Not applicable. #### Representation - **Present:** Not applicable. - **Improve:** Not applicable. #### Psychological safety - **Present:** Not applicable. - **Improve:** Not applicable. ## ✅ Logical Validity #### Strong reasoning - **Problem-solution fit.** Long agent outputs bury important details; an observer that surfaces them addresses that directly. #### Fallacies - **Unsupported benefit claim** — Location: "to help keep you in the loop." Weakness: no evidence of what it catches or how often. Fix: show an example flag and a false-positive rate. #### Tone of voice - Product-announcement casual, concise. ## 🛡️ DARVO Analysis - **Deny.** Not present. - **Attack.** Not present. - **Reverse victim/offender.** Not present. - **Overall.** Not applicable to this source. ## 🎯 Conclusion - **Bottom line.** A built-in Claude Code mod that runs a side agent to flag important details in Claude's output; also a first-party showcase for the mods API. - **Next steps.** Enable it for a week and note what it flags; read the mods tutorial before writing a custom observer (e.g., for WordPress security patterns). ## 🔄 Refactor Appendix — We're adding a new plugin to Claude Code: You should Know ### 📚 Literature Atoms #### [H1] A side agent flags what Claude's output buries > [!quote] ID:rw_01m3zdb76m520cs19stmv2acjp > It scans Claude's output for important information you might miss to help keep you in the loop. **Context & use**: - Post 1, second line. - The supervision pattern: a second agent watches the first for the human. **Tags**: #xref **Highlight**: inline (Phase 2) <!-- atomicity: 1 idea ✓ --> #### [H2] Mods spin off observer sideagents > [!quote] ID:rw_01m3zdb76m520cs19stmv2acjp > "You should know" is a mod that works by spinning off a sideagent to observe Claude's output. **Context & use**: - Post 2. - Links the plugin to the mods API. **Tags**: #xref #articleseed **Highlight**: inline (Phase 2) <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates #### [P1] an-observer-agent-fixes-attention-not-correctness **Claim** (one sentence, Courtney's voice): A second agent watching the first helps the human notice what was said, but it doesn't make what was said true. **Evidence pointer**: [[#H1]] **Scope & conditions**: Holds for summarizing/flagging observers; a verifying observer that runs tests is a different tool. **Links**: - [[Using Claude Code Spending Your Effort]] — because extends agent supervision and verification - [[I Was Using Fable Wrong, This Is How I Fixed It]] — because applies-in Claude Code workflow habits <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Sketch a WordPress-security observer mod (unescaped echo, missing nonce, raw $wpdb) using the mods API #fleeting #inbox ### → Things 3 - [ ] Enable You Should Know in Claude Code and note what it flags for a week — ClaudeDevs post #things <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Tweets/We're Adding a New Plugin to Claude Code You Should....md|We're Adding a New Plugin to Claude Code: You Should... by ClaudeDevs]] ## Highlights - It scans Claude's output for important information you might miss to help keep you in the loop. ([View Highlight](https://read.readwise.io/read/01m43ftgk67zj752gw8jd9jh2z)) - Note: [H1] A side agent flags what Claude's output buries - Tags: atom/literature/h1 - "You should know" is a mod that works by spinning off a sideagent to observe Claude's output. ([View Highlight](https://read.readwise.io/read/01m43ftpfjxj1fzqs76a7s56qp)) - Note: [H2] Mods spin off observer sideagents - Tags: atom/literature/h2