Mods let you hook into Claude Code while it works: block a command, rewrite a tool call, redact a secret from a result, or draw a live pane above the prompt. Until recently that meant writing the module yourself, packaging it as a plugin and publishing a marketplace. The Mods catalog on aitmpl.com removes those steps: 38 mods, one command each.
A Mod is a Claude Code plugin whose behavior lives in a function-hooks module. A shell hook reacts to an event after the fact. A function hook wraps the event: your function receives the engine object $, the event e, and a next continuation, and decides what happens.
That gives three positions instead of one:
next(e) and inspect what came back.next({ ...e, input }) with changed data, or change the result on the way out.{ deny: "reason" } and never call next. The tool does not run.
The module runs in a sandbox of its own, with no DOM and no Node. Everything outside it goes through $: $.ui, $.http, $.fs, $.process, $.state, $.store, $.session, $.command, $.tool, $.model and $.clock. That is what makes a mod more than a filter.
Mods are distributed as plugins. The standard route is a plugin marketplace (a repo with .claude-plugin/marketplace.json):
/plugin marketplace add <repo>
/plugin install <mod>@<marketplace>
/reload-plugins
With aitmpl.com you skip that. Pick a mod on the site, copy the command, run it in your project:
npx claude-code-templates@latest --mod security/secret-redactor
The CLI downloads the whole plugin directory (.claude-plugin/plugin.json, hooks/, commands, types, tests) and writes it to .claude/skills/secret-redactor/. Claude Code loads it on the next session as secret-redactor@skills-dir.
Two details that trip people up:
.claude/skills/ is only read once you have accepted the workspace trust prompt. In a folder you never opened interactively (for example a headless claude -p run) the mod does not load. Install it in ~/.claude/skills/ if you want it everywhere.userConfig in plugin.json and read from your
{ "pluginConfigs": { "npm-to-pnpm-rewriter@skills-dir": { "options": { "manager": "yarn" } } } }
Use the wrong key ("npm-to-pnpm-rewriter") and every option silently stays at its default.
Seven categories, 38 mods:
secret-redactor, block-destructive-commands, protected-paths-guard, large-edit-confirmation, jev-auto-mode, jev-guardrails, jev-vercel-sandbox
aitmpl (a component browser inside Claude Code), npm-to-pnpm-rewriter, webfetch-cache, jev-model-router, jev-skill-suggestion
agent-flow, git-sidebar, tool-timing-badge
prompt-cache-control, universal-audit-log
linear-tickets, vercel-deploys, neon-branch-per-session, websearch-to-exa, pi-agent-for-claude
admin-capability-lockdown
snake and tetris to doom, pacman and tool-defense, where every enemy is one of Claude's real tool calls
Mods are not terminal-only. The official guide demos each one in both Claude Code in the terminal and Claude Code Desktop, and surfaces like panes and status lines are drawn through $.ui.resolve(e), which returns the constructors for whichever surface you are on. Same plugin, same install command, both places.
CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1.)$ API is still young and can change between releases. The catalog's type definitions follow Claude Code 2.1.283 and every mod is typechecked against them in CI.
Before touching the code, check in this order:
$.ui.log goes to the transcript and to ~/.claude/debug/<session-id>.txt; in pluginConfigs keyed by name@skills-dir? claude --debug show hooks module <name>@skills-dir loaded?
A note on model routers: a mod that rewrites model or effort at turn.step changes that request only. The header, the status line and the effort box show your session's setting and will not move.
npx claude-code-templates@latest --mod games/pacman
Browse the catalog at aitmpl.com, validate any installed mod with claude plugin validate <dir>, and run its tests with claude plugin test <dir>.
Every mod has its own page with the install command, the options it accepts and what it hooks.