Claude Code mods: what 1,017 community mods can reach

Two days after Anthropic launched Claude Mods, a community catalogue lists more than a thousand of them. The useful part is not the count but what the catalogue records for each mod: what it can read, write, run or send over the network. This article covers what the catalogue is, what its numbers show, what people are building, and what to check before you install one. We read the catalogue’s data and Anthropic’s documentation; we did not run any mod.

What the catalogue is

Awesome Claude Code Mods (with a searchable website) is a community project, not an Anthropic one. Its repository dates from September 15, before the launch, and its README refers to an earlier list “from before the rename” (the feature was called function hooks), so not all of these mods appeared in two days. A scheduled scan finds public mod repositories on GitHub and checks each mod’s source with claude plugin validate. It then sorts mods into reach levels by the widest thing their $ calls can touch:

• L0, draws and remembers: only changes the interface and keeps its own state.
• L1, reads: can read files or other local data.
• L2, writes or runs: can write files, run processes or drive Claude.
• L3, network: can reach the network.

The catalogue says this is “a footprint, not a verdict”: a pull-request tracker has to run gh, so being at L2 does not make it bad. It also says validation checks the source without running the mod and does not guarantee safe behavior.

The numbers

The scan data, generated on October 3 against Claude Code 2.1.288, counts 1,017 mods.

Reach level	Mods
L0, draws and remembers	112
L1, reads	220
L2, writes or runs	518
L3, network	167

By the catalogue’s own validation, 823 mods passed, 166 passed with warnings and 28 failed on 2.1.288. Counting the data file ourselves, the mods come from about 530 authors in about 650 repositories.

The data also lists what each mod can do. Across the 1,017, 409 run processes, 381 read environment variables, 379 read files, 325 drive Claude and 150 reach the network. The 167 at L3 and the 150 labelled “network” come from different groupings in the data, and we report both as published. On what they observe, 322 see every tool call and 308 see every prompt. Environment variables can hold API keys, and a mod that sees every prompt sees what you type. These are capabilities a mod’s code uses, not evidence that any of them misuses them, and the scanner does not say what a mod does with what it reads.

One caution about the totals: mods are a new feature, and the catalogue notes it discovers repositories through GitHub’s index. Treat the figures as a snapshot that will move daily.

What people are building

The catalogue’s curated lists group mods by purpose:

• Dashboards and usage: cctop, a btop-style pane with context fill, tokens, cost and rate limits; quota and spend meters above the prompt.
• Git and CI: cc-pr-tracker, which shows watched pull requests with review and check state above the prompt; mods that pin GitHub Actions runs or Vercel deploy status.
• Safety and privacy: secret-redactor, which swaps secrets, email addresses and IPs for placeholders before the model reads them; launch-codes, which asks for a one-time code before risky commands such as a force push.
• Memory and context: segmented memory stored in one SQLite file, and aside, a read-only side chat about the session so far.
• While you wait: games, pets and quotes shown above the prompt while a turn runs.

Two entries in the “while you wait” list show why reach matters. According to the catalogue, a pixel-pet mod (nibbl) syncs event types and times to its own server, and a Doom mod (intermission) downloads and runs a game and connects to a shared game server. Both are described openly in the catalogue, but neither is something you would guess from the mod’s name alone.

What Anthropic says

Anthropic’s launch post says mods run with the same access to your machine as Claude Code, are not sandboxed, and should be installed only from sources you trust, as with any code. On Team and Enterprise plans, a built-in sec-default mod loads first so that user-installed mods cannot override permission rules; its source is public beside the other three built-in mods (diff, agents-md, telemetry). We could not find Anthropic documentation for other organization-level switches by name, so we do not list any.

Before you install a mod

These are our suggestions, built on the sources above:

• Read the repository. Anthropic’s own tutorial says to, and it is the only check that sees what the code does.
• Check the access level. Look the mod up in the catalogue. A status line at L0 and a mod at L3 are different decisions.
• Treat a guard as a safety net. Anthropic’s tutorial says its Blast Radius example reads command text, so use permission rules for hard blocks.
• Install one at a time, and use claude --safe-mode (documented in the changelog since 2.1.169, it starts Claude Code with plugins and other customizations off) if something behaves oddly.
• Run unknown mods away from valuable data, in a separate user account or a VM, as with any agent you are testing.

What changed in plugins this week

The 2.1.285 to 2.1.288 release notes add claude plugin configure and claude plugin install --config so a plugin’s options can be set from the command line, a $.ui.selection() call for mods, and a group of plugin fixes, including plugin LSP servers receiving unsubstituted placeholders and git-subdir installs failing on git older than 2.39.

The release notes also list a built-in mod called “You should know,” which Anthropic describes as a side agent that watches your back and flags things you or Claude might miss. You enable it with /plugin enable cc-plugin-you-should-know@builtin, and the notes say it is for first-party sessions with telemetry on. A post from Anthropic’s official ClaudeDevs account on X, which the editor showed us as a screenshot (we did not open the post ourselves), describes it as a mod that works by spinning off a side agent to observe Claude’s output. A screenshot from the same thread appears to show it in use: under Claude’s summary of changes to a checkout and payments feature, a “Heads up” note warns that automatic retries could charge a customer twice and offers the choices “Learn more,” “Knew this already” and “Dismiss.” A GitHub issue opened on October 2 reports that Claude Code’s startup tip for it fails with “Plugin is not installed” on 2.1.287. We have not used it and found no further documentation beyond these.

What to watch

• Catalogue numbers. They move daily and depend on GitHub discovery.
• Directory submissions. Anthropic’s tutorial says the Claude directory accepts plugins that include mods, so listings there may become the main place to find them.
• Independent review. The scan reads source for capabilities. It does not say what a mod does with them, and no one reviews each mod’s behavior.

Editorial review: October 3, 2026 (AI). Prepared from the catalogue’s README, catalogue notes and data file (generated October 3, 2026, scanned against Claude Code 2.1.288), Anthropic’s launch post and tutorial, the Claude Code changelog, the public mods README and the GitHub issue cited, read the same day. The ClaudeDevs post and thread about “You should know” were shown to us by the editor as screenshots; we did not open them ourselves. Counts of authors, repositories and capability labels are our tallies of the catalogue’s data file. We did not install, run or review any mod. The suggestions in “Before you install a mod” are ours.