A collection of Claude Code plugins for enhanced productivity and workflow automation.

Add this marketplace to your Claude Code installation:

/plugin marketplace add bromanko/llm-agents

/plugin marketplace add /path/to/this/repo

Tools and commands for working with Jujutsu (jj) version control.

Context tip: Always use --color=never (or --git for diffs) when running jj commands via Bash. ANSI escape codes waste 2-3x the tokens.

Commands:

• /jj:commit - Analyze jj status and create logical commits with descriptive messages
• /jj:workspace-cleanup - Clean up empty jujutsu workspaces

Configuration:

• Global: ~/.pi/agent/jj-commit.json
• Project: .pi/jj-commit.json
• Project config overrides global config

Example:

{
  "model": "dbx-bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0"
}

The configured provider/id must match the exact model shown by pi's model selector.

Hooks:

• Blocks mutating git commands in jujutsu repositories, guiding you to jj equivalents

Installation:

/plugin install jj@bromanko-llm-agents

AI-assisted Git commit workflow with optional hunk-level split commits.

Commands:

• /git-commit - Analyze Git changes, propose commit messages, and split mixed changes by file or hunk

Configuration:

• Global: ~/.pi/agent/git-commit.json
• Project: .pi/git-commit.json
• Project config overrides global config

Example:

{
  "model": "anthropic/claude-opus-4-1"
}

The configured provider/id must match the exact model shown by pi's model selector.

Installation:

/plugin install git@bromanko-llm-agents

Comprehensive word exploration with dictionary definitions, etymology, and literary examples.

Skills:

• word-explorer - Provides word profiles combining Webster's 1913 definitions, literary quotations, and usage guidance

Installation:

/plugin install word-explorer@bromanko-llm-agents

Master typed functional programming in TypeScript with fp-ts library guidance.

Skills:

• fp-ts - Expert guidance for Option, Either, Task, TaskEither, and functional composition patterns

Installation:

/plugin install fp-ts@bromanko-llm-agents

Gleam-specific code review skills for quality, security, performance, and testing.

Skills:

• code-review - Quality and idiom analysis for Gleam code
• security-review - Security audit focusing on FFI safety, input validation, and dependencies
• performance-review - BEAM runtime optimization analysis
• test-review - Test coverage and quality assessment

Installation:

/plugin install gleam-review@bromanko-llm-agents

F#-specific code review skills for quality, security, performance, and testing.

Skills:

• code-review - Quality and idiom analysis for F# code
• security-review - Security audit focusing on .NET interop, serialization, input validation, and dependencies
• performance-review - .NET runtime optimization analysis (allocations, async, collections)
• test-review - Test coverage and quality assessment (Expecto, xUnit, FsCheck)

Installation:

/plugin install fsharp-review@bromanko-llm-agents

Elm-specific code review skills for quality, security, performance, and testing.

Skills:

• code-review - Quality and idiom analysis for Elm code (TEA, type design, decoders)
• security-review - Security audit focusing on port safety, JSON validation, and XSS prevention
• performance-review - Virtual DOM rendering and data structure optimization analysis
• test-review - Test coverage and quality assessment (elm-test, fuzz testing, Test.Html)

Installation:

/plugin install elm-review@bromanko-llm-agents

The repository includes an interactive code review command in pi/code-review/extensions/index.ts.

/review <language> [types...] [-r|--revisions <range>] [--fix <high|medium|low|all>] [--report <high|medium|low|all>]

• language selects the review skill family (gleam, fsharp, elm, etc.).
• types optionally narrows to code, security, performance, and/or test.
• -r/--revisions selects what changes to review. Default is @.
• --fix auto-queues follow-up fix requests at or above the given severity threshold.
• --report renders a deterministic markdown report for findings at or above the given severity threshold. Use it with pi -p.

/review gleam
/review gleam -r main..@
/review gleam code security -r abc123
/review gleam -r @ --fix high
/review fsharp test -r main..@ --fix medium
pi -p "/review gleam --report all"
pi -p "/review fsharp security --report medium"

/review gathers code changes by trying:

1. jj diff -r <range> --git
2. git fallback (git diff for ranges, git show --patch for single revisions)

If both fail, the command reports a deterministic error including the range and failing commands.

After installing a plugin:

1. Browse commands: Use / to see available commands
2. Get help: Most commands support /command --help
3. Configure: Check plugin documentation for configuration options

1. Create a new directory under plugins/
2. Add a plugin.json manifest file
3. Add your plugin components (commands, agents, hooks, etc.)
4. Update .claude-plugin/marketplace.json to register the plugin
5. Submit a pull request

plugins/your-plugin/
├── plugin.json          # Plugin manifest
├── commands/            # Custom commands (optional)
├── agents/              # Custom agents (optional)
└── README.md           # Plugin documentation (optional)

1. Clone this repository
2. Add the local marketplace:

   /plugin marketplace add ./path/to/clone

3. Install plugins for testing:

   /plugin install plugin-name@bromanko-llm-agents

Validate the marketplace configuration:

claude plugin validate .

Run the full test suite from the repository root:

npm test

Run a single test file:

node --experimental-strip-types --test path/to/file.test.ts

Test files are co-located with the module under test and named <module>.test.ts (or .test.js for existing JavaScript modules).

Exception: extension directories. Never place test files or non-extension support modules inside directories registered as pi.extensions in package.json (e.g. pi/jj/extensions/, pi/web/extensions/). Pi auto-discovers every *.ts file in those directories and loads them as extensions — test files will execute their test() calls on import, and library modules without a default export will fail to load. Place extension tests in test/extensions/ and shared library code in a package-local lib/ directory (for example pi/web/lib/).

For extension tests, use the shared mock helper in test/helpers.ts:

import test from "node:test";
import assert from "node:assert/strict";
import { createMockExtensionAPI } from "../../test/helpers.ts";
import extension from "./my-extension.ts";

test("registers and handles tool_call", async () => {
  const pi = createMockExtensionAPI();
  extension(pi as any);

  const [handler] = pi.getHandlers("tool_call");
  const result = await handler(
    { toolName: "bash", input: { command: "git push" } },
    { cwd: process.cwd(), sessionManager: { getBranch: () => [] } },
  );

  assert.equal(result?.block, true);
});

No external testing dependencies are required. Tests use Node.js built-ins (node:test, node:assert/strict) and TypeScript support from --experimental-strip-types.

This repository now includes a first-party read override at pi/files/extensions/read.ts.

• Purpose: keep the familiar read tool while adding direct line targeting for common paging workflows.
• Supported parameters: path, offset, limit, endLine, tail, aroundLine, and context.
• Examples:
  • read package.json endLine=5
  • read package.json tail=5
  • read README.md aroundLine=280 context=2
• Image behavior: known image extensions (.png, .jpg, .jpeg, .gif, .webp) still delegate to pi's built-in image read pipeline.
• Prompt guidance: prefer endLine, tail, and aroundLine over head, tail, or sed when you need a targeted read.

Run pi with this extension directly:

pi -e ./pi/files/extensions/read.ts

This repository now includes a first-party fetch extension at pi/web/extensions/fetch.ts.

• Purpose: fetch HTTP(S) URLs and return model-readable output with metadata (URL, optional Final URL, Status, Content-Type, Method) plus a readable body.
• MVP content handling:
  • application/json → pretty-printed JSON
  • text/html → conservative HTML-to-text conversion
  • text/* and markdown-like text → passthrough text
  • other content types → best-effort text fallback
• Truncation behavior: output is bounded by line/byte limits (defaults: 2000 lines, 50KB). When truncated, full output is saved to a temp file and the response includes a truncation notice with the saved path.
• Safety: non-HTTP(S) schemes are rejected (for example file://).
• Known non-goals (deferred): binary/PDF conversion and site-specific advanced scraping pipelines.

Run pi with this extension directly:

pi -e ./pi/web/extensions/fetch.ts

This repository now includes a first-party web_search extension at pi/web/extensions/web-search/index.ts.

• Purpose: discover current web sources via Brave Search and return bounded, citation-friendly output.
• Provider scope (v1): brave only.
  • provider: "auto" resolves to Brave when BRAVE_API_KEY is set.
  • Anthropic/Codex provider-native adapters are intentionally deferred to v2.
• Required env var: BRAVE_API_KEY
• Parameters:
  • query (required)
  • provider (auto | brave)
  • recency (day | week | month | year)
  • limit (clamped to 1..10, default 5)
  • enrich (boolean, default false)
  • fetchTop (clamped to 0..5, default 0)
• Optional enrichment: when enrich=true, top results are fetched through the local fetch pipeline (pi/web/lib/fetch-core.ts) and appended as bounded excerpts.

Run pi with this extension directly:

pi -e ./pi/web/extensions/web-search/index.ts

This repository includes a universal LSP extension at pi/lsp/extensions/lsp.ts that provides automatic diagnostics and code intelligence for any language with a configured LSP server.

• Auto diagnostics on write/edit: after every write or edit tool call, the extension sends the file to the appropriate LSP server and appends any diagnostics directly to the tool result. No explicit tool call needed.
• Format on write: optionally reformats the file after write/edit using the LSP server's formatting capability. Enabled by default.
• lsp tool for code intelligence: a single tool with an action parameter for on-demand queries:
  • languages — list detected servers and their status
  • diagnostics — get current diagnostics for a file
  • definition — go to definition
  • references — find all references
  • hover — get hover/type information
  • symbols — document or workspace symbols (pass query for workspace)
  • rename — rename a symbol
  • code_actions — list available code actions
  • incoming_calls / outgoing_calls — call hierarchy

The extension merges configuration from three layers (lowest to highest precedence):

1. Built-in defaults (pi/lsp/lib/defaults.json)
2. User config (~/.pi/agent/lsp.json)
3. Project config (.pi/lsp.json)

Example .pi/lsp.json:

{
  "formatOnWrite": true,
  "diagnosticsOnWrite": true,
  "autoCodeActions": false,
  "idleTimeoutMinutes": 10,
  "servers": {
    "typescript-language-server": {
      "args": ["--stdio", "--log-level=1"]
    },
    "custom-server": {
      "command": "my-lsp",
      "args": ["--stdio"],
      "fileTypes": [".custom"],
      "rootMarkers": ["custom.config"]
    }
  }
}

• TypeScript/JavaScript via typescript-language-server (.ts, .tsx, .js, .jsx)

Additional servers can be added via config files.

The extension is auto-loaded when using this repository as a pi package. To run directly:

pi -e ./pi/lsp/extensions/lsp.ts

Integration tests require typescript-language-server and typescript in PATH:

nix develop .#lsp-test -c node --experimental-strip-types --test pi/lsp/test/integration/typescript.e2e.test.ts

Cross-session token usage and cost reporting.

/session-stats [range] [by day|project|model]

/session-stats
/session-stats today
/session-stats last 7 days
/session-stats this month by project
/session-stats 2026-04-01..2026-04-06 by day
/session-stats all time by model

• Scans all saved sessions across all projects on this machine.
• Only saved sessions are counted. Ephemeral --no-session runs are invisible.
• Ranges are interpreted in local time.
• Does not invoke a model. Deterministic and read-only.

MIT