coderlm 0.1.1__tar.gz

coderlm-0.1.1/.claude/CLAUDE.md

@@ -0,0 +1,61 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## What This Is
+
+A CLI that implements the RLM (Recursive Language Model) pattern: instead of feeding all files into an LLM's context, give it a file listing and let it use tools to peek, decompose, and recursively call itself on subsets. Ships as both an npm package (`bunx coderlm`) and a PyPI package (`uvx coderlm`).
+
+## Commands
+
+```bash
+bun test                          # run tests
+bun test --watch                  # run tests in watch mode
+bun test --filter "pattern"       # run a single test by name
+just publish                      # bump patch + publish to npm and pypi
+just publish npm minor            # bump minor + publish to npm only
+just publish pypi major           # bump major + publish to pypi only
+bun x ultracite fix               # format and lint
+bun x ultracite check             # check for issues
+```
+
+## Architecture
+
+The entire CLI is a single bash script at `src/coderlm`. It:
+1. Parses args (command, globs, --prompt, --max-depth, --dry-run)
+2. Expands globs via `fd` (falls back to `find`)
+3. Builds a system prompt with the file list and RLM instructions
+4. Dispatches to the appropriate agent with agent-specific flags:
+   - **claude**: `-p --append-system-prompt <sys> --allowedTools Bash <prompt>` (system prompt separate)
+   - **codex**: `exec --full-auto <combined>` (system + task combined)
+   - **gemini**: `-p <combined> --yolo` (command is word-split, e.g. `bunx --bun @google/gemini-cli`)
+   - **generic**: `<combined>` as single arg
+
+The Python package (`src/__init__.py`) is a thin wrapper that `os.execvp`s the bash script.
+
+**Context guards** (`src/bashrlm.sh`): Wraps common high-output commands (`cat`, `rg`, `grep`, `jq`, `find`, etc.) with truncation to prevent agents from blowing up their context windows. Injected into every non-interactive bash subshell the agent spawns via `BASH_ENV`. The file `src/bashrlm.md` contains the corresponding instructions that get appended to the system prompt, describing the truncation behavior and rules to the agent.
+
+**Model passthrough**: The agent command is word-split (`read -ra _agent_cmd <<< "$agent"`), so model flags can be passed inline:
+```bash
+coderlm "claude --model claude-haiku-4-5" "**/*.ts" --prompt "..."
+coderlm "codex -m o4-mini" "**/*.py" --prompt "..."
+```
+
+## Testing
+
+Unit tests are in `src/unit.test.ts` and integration tests in `src/integration.test.ts`, using `bun:test`. Unit tests use `--dry-run` which prints the constructed command as null-delimited args to stdout instead of exec-ing. This lets tests verify argument construction for each agent without actually running them.
+
+## Known Issues
+
+See `.claude/PROBLEMS.md` for a running log of problems and solutions (e.g., Claude Code nesting limitations, MCP server conflicts).
+
+## Dual Publishing
+
+Versions are kept in sync across `package.json` and `pyproject.toml`. The justfile `_bump` recipe uses `npm version` then syncs to pyproject.toml via perl.
+
+
+# Code Standards
+
+This project uses **Ultracite** (Biome-backed). Run `bun x ultracite fix` before committing.
+
+TypeScript (test files only): use `const` by default, `async/await` over promise chains, explicit types where clarity is improved, `unknown` over `any`. Don't use `.only` or `.skip` in committed tests.

coderlm-0.1.1/.claude/PROBLEMS.md

@@ -0,0 +1,31 @@
+# Problems & Solutions
+
+Running log of known issues and workarounds.
+
+## Claude Code cannot be spawned from within a Claude Code session
+
+**Problem**: Running `coderlm claude ...` from inside a Claude Code session (or any child process of Claude Code) fails silently or with "cannot be launched inside another Claude Code session". This happens even after unsetting `CLAUDECODE` — the nesting guard goes deeper than just the env var (likely shared runtime resources or lock files).
+
+**Impact**: Cannot integration-test the `claude` agent path from within Claude Code. The codex and gemini paths work fine.
+
+**Workaround**: Test the claude path from a standalone terminal:
+```bash
+./src/coderlm claude "**/*.ts" --prompt "Summarize" --max-depth 1
+```
+
+The integration test suite uses `--dry-run` for claude to verify command construction without actually spawning a session.
+
+## CLAUDECODE env var leaks into macOS session
+
+**Problem**: Claude Code sets `CLAUDECODE` in child processes. If something triggers `launchctl setenv CLAUDECODE ...`, it persists across all new terminals system-wide — even standalone ones outside the IDE.
+
+**Fix**:
+```bash
+launchctl unsetenv CLAUDECODE
+```
+
+## Gemini CLI hangs when MCP servers conflict
+
+**Problem**: When running `coderlm` with Gemini from inside another agent session (e.g., Claude Code), the shared `nia` MCP server can't handle concurrent connections. Gemini hangs during startup trying to connect.
+
+**Workaround**: Pass `--allowed-mcp-server-names none` to disable MCP servers, or run from a standalone terminal. The integration test sets `GEMINI_ALLOWED_MCP_SERVERS=none` in the env.

coderlm-0.1.1/.claude/settings.json

@@ -0,0 +1,15 @@
+{
+  "hooks": {
+    "PostToolUse": [
+      {
+        "matcher": "Write|Edit",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "just fmt"
+          }
+        ]
+      }
+    ]
+  }
+}

coderlm-0.1.1/.github/workflows/test.yml

@@ -0,0 +1,16 @@
+name: Test
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: oven-sh/setup-bun@v2
+      - run: bun install --frozen-lockfile
+      - run: bun test src/unit.test.ts

coderlm-0.1.1/.gitignore

@@ -0,0 +1,38 @@
+# dependencies (bun install)
+node_modules
+
+# output
+out
+dist
+*.tgz
+
+# python
+__pycache__
+*.egg-info
+
+# code coverage
+coverage
+*.lcov
+
+# logs
+logs
+_.log
+report.[0-9]_.[0-9]_.[0-9]_.[0-9]_.json
+
+# dotenv environment variable files
+.env
+.env.development.local
+.env.test.local
+.env.production.local
+.env.local
+
+# caches
+.eslintcache
+.cache
+*.tsbuildinfo
+
+# IntelliJ based IDEs
+.idea
+
+# Finder (MacOS) folder config
+.DS_Store

coderlm-0.1.1/.mise.toml

@@ -0,0 +1,4 @@
+[tools]
+bun = "1"
+just = "1.40"
+python = "3.13"

coderlm-0.1.1/.zed/settings.json

@@ -0,0 +1,62 @@
+{
+  "file_scan_exclusions": [
+    "**/.DS_Store",
+    "**/.git",
+    "**/.openai_test_cache/",
+    "**/.pytest_cache/",
+    "**/.ruff_cache",
+    "**/.tsbuildinfo",
+    "**/.venv/",
+    "**/__pycache__",
+    "**/node_modules/",
+    "dist/**"
+  ],
+  "formatter": "language_server",
+  "format_on_save": "on",
+  "lsp": {
+    "typescript-language-server": {
+      "settings": {
+        "typescript": {
+          "preferences": {
+            "includePackageJsonAutoImports": "on"
+          }
+        }
+      }
+    }
+  },
+  "languages": {
+    "JavaScript": {
+      "formatter": {
+        "language_server": {
+          "name": "biome"
+        }
+      },
+      "code_actions_on_format": {
+        "source.fixAll.biome": true,
+        "source.organizeImports.biome": true
+      }
+    },
+    "TypeScript": {
+      "formatter": {
+        "language_server": {
+          "name": "biome"
+        }
+      },
+      "code_actions_on_format": {
+        "source.fixAll.biome": true,
+        "source.organizeImports.biome": true
+      }
+    },
+    "TSX": {
+      "formatter": {
+        "language_server": {
+          "name": "biome"
+        }
+      },
+      "code_actions_on_format": {
+        "source.fixAll.biome": true,
+        "source.organizeImports.biome": true
+      }
+    }
+  }
+}

coderlm-0.1.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 zenbase-ai
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

coderlm-0.1.1/PKG-INFO

@@ -0,0 +1,92 @@
+Metadata-Version: 2.4
+Name: coderlm
+Version: 0.1.1
+Summary: Process large codebases using the RLM (Recursive Language Model) pattern
+License-Expression: MIT
+License-File: LICENSE
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+
+# coderlm
+
+Process large codebases with Coding Agents using the **RLM (Recursive Language Model)** pattern — with bundled context guards to keep the agent's output from flooding its own context window.
+
+Instead of stuffing all files into an LLM's context window, give it a file listing and let it use tools to peek, decompose, and recursively call itself on subsets. This keeps each agent focused on a manageable scope while covering arbitrarily large codebases.
+
+## Install
+
+```bash
+# npm / bun
+bunx coderlm@latest
+npx coderlm@latest
+
+# pypi
+uvx coderlm
+pipx run coderlm
+```
+
+Or install globally:
+
+```bash
+npm i -g coderlm    # npm
+uv tool install coderlm  # pypi
+```
+
+## Usage
+
+```
+coderlm <agent> <globs...> --prompt "<task>" [--max-depth N] [--allowedTools TOOLS]
+```
+
+### Examples
+
+```bash
+# Codex
+coderlm codex "src/**/*.ts" --prompt "Find all TODO comments"
+coderlm codex "src/**" "lib/**" "test/**" --prompt "Find dead code"
+coderlm codex "**/*.ts" --prompt "Summarize the codebase" --max-depth 2
+
+# Gemini
+coderlm "bunx --bun @google/gemini-cli" "**/*.py" --prompt "Review for security issues"
+coderlm "bunx --bun @google/gemini-cli" "src/**" --prompt "Architecture overview"
+
+# Claude (non-recursive only — Claude cannot spawn nested Claude sessions)
+coderlm claude "src/**" --prompt "Fix type errors" --allowedTools "Bash,Edit"
+```
+
+## How It Works
+
+1. **Expand globs** into a file listing using `fd` (or `find` as fallback)
+2. **Build a system prompt** containing the file list and RLM instructions (explore, decompose, aggregate)
+3. **Inject context guards** via `BASH_ENV` so every bash subshell the agent spawns has output truncation active
+4. **Launch the agent** with agent-specific flags for non-interactive execution
+
+The agent receives a file listing — not file contents. It uses shell tools (`rg`, `cat`, `head`, `jq`, etc.) to inspect files as needed. For large file sets (>20 files), it spawns recursive sub-agents on subsets.
+
+### Context Guards (bundled)
+
+`bashrlm.sh` is bundled and auto-activates for every agent run. It wraps high-output commands (`cat`, `grep`, `rg`, `jq`, `find`, `ls`, `curl`, etc.) with automatic truncation, preventing the agent from flooding its own context window with oversized output.
+
+Truncation uses head+tail mode — the agent sees the start and end of any large output, with the middle omitted:
+
+```
+[TRUNCATED — showing 2000 of 15000 chars, first and last 1000]
+```
+
+Guards are redirect-aware: piping to a file (`> /tmp/out.txt`) bypasses truncation, so multi-step processing works naturally. The agent's system prompt includes instructions for this pattern.
+
+### Supported Agents
+
+| Agent         | Command                         | Notes                                                              |
+| ------------- | ------------------------------- | ------------------------------------------------------------------ |
+| OpenAI Codex  | `codex`                         | Recommended — supports recursive sub-agents                        |
+| Google Gemini | `bunx --bun @google/gemini-cli` | Supports recursive sub-agents                                      |
+| Claude Code   | `claude`                        | Non-recursive only — cannot spawn nested Claude sessions           |
+| Any CLI       | `my-agent`                      | Combined prompt passed as single argument                          |
+
+## When to Use
+
+- Many files to analyze (>10)
+- Task benefits from divide-and-conquer
+- Total content exceeds a comfortable context window
+- You want the agent to explore strategically rather than read everything upfront

coderlm-0.1.1/README.md

@@ -0,0 +1,83 @@
+# coderlm
+
+Process large codebases with Coding Agents using the **RLM (Recursive Language Model)** pattern — with bundled context guards to keep the agent's output from flooding its own context window.
+
+Instead of stuffing all files into an LLM's context window, give it a file listing and let it use tools to peek, decompose, and recursively call itself on subsets. This keeps each agent focused on a manageable scope while covering arbitrarily large codebases.
+
+## Install
+
+```bash
+# npm / bun
+bunx coderlm@latest
+npx coderlm@latest
+
+# pypi
+uvx coderlm
+pipx run coderlm
+```
+
+Or install globally:
+
+```bash
+npm i -g coderlm    # npm
+uv tool install coderlm  # pypi
+```
+
+## Usage
+
+```
+coderlm <agent> <globs...> --prompt "<task>" [--max-depth N] [--allowedTools TOOLS]
+```
+
+### Examples
+
+```bash
+# Codex
+coderlm codex "src/**/*.ts" --prompt "Find all TODO comments"
+coderlm codex "src/**" "lib/**" "test/**" --prompt "Find dead code"
+coderlm codex "**/*.ts" --prompt "Summarize the codebase" --max-depth 2
+
+# Gemini
+coderlm "bunx --bun @google/gemini-cli" "**/*.py" --prompt "Review for security issues"
+coderlm "bunx --bun @google/gemini-cli" "src/**" --prompt "Architecture overview"
+
+# Claude (non-recursive only — Claude cannot spawn nested Claude sessions)
+coderlm claude "src/**" --prompt "Fix type errors" --allowedTools "Bash,Edit"
+```
+
+## How It Works
+
+1. **Expand globs** into a file listing using `fd` (or `find` as fallback)
+2. **Build a system prompt** containing the file list and RLM instructions (explore, decompose, aggregate)
+3. **Inject context guards** via `BASH_ENV` so every bash subshell the agent spawns has output truncation active
+4. **Launch the agent** with agent-specific flags for non-interactive execution
+
+The agent receives a file listing — not file contents. It uses shell tools (`rg`, `cat`, `head`, `jq`, etc.) to inspect files as needed. For large file sets (>20 files), it spawns recursive sub-agents on subsets.
+
+### Context Guards (bundled)
+
+`bashrlm.sh` is bundled and auto-activates for every agent run. It wraps high-output commands (`cat`, `grep`, `rg`, `jq`, `find`, `ls`, `curl`, etc.) with automatic truncation, preventing the agent from flooding its own context window with oversized output.
+
+Truncation uses head+tail mode — the agent sees the start and end of any large output, with the middle omitted:
+
+```
+[TRUNCATED — showing 2000 of 15000 chars, first and last 1000]
+```
+
+Guards are redirect-aware: piping to a file (`> /tmp/out.txt`) bypasses truncation, so multi-step processing works naturally. The agent's system prompt includes instructions for this pattern.
+
+### Supported Agents
+
+| Agent         | Command                         | Notes                                                              |
+| ------------- | ------------------------------- | ------------------------------------------------------------------ |
+| OpenAI Codex  | `codex`                         | Recommended — supports recursive sub-agents                        |
+| Google Gemini | `bunx --bun @google/gemini-cli` | Supports recursive sub-agents                                      |
+| Claude Code   | `claude`                        | Non-recursive only — cannot spawn nested Claude sessions           |
+| Any CLI       | `my-agent`                      | Combined prompt passed as single argument                          |
+
+## When to Use
+
+- Many files to analyze (>10)
+- Task benefits from divide-and-conquer
+- Total content exceeds a comfortable context window
+- You want the agent to explore strategically rather than read everything upfront

coderlm-0.1.1/SKILL.md

@@ -0,0 +1,20 @@
+---
+name: coderlm
+description: Use when a task involves many files (>10), the total content exceeds comfortable context size, or the task benefits from divide-and-conquer. Runs an agent with a file listing instead of file contents, letting it peek strategically and recursively decompose into sub-agents. Bundles bashrlm context guards — output from high-output commands (cat, grep, rg, jq, find, ls, curl, etc.) is automatically truncated so the agent never floods its own context window.
+---
+
+## Usage
+
+```
+coderlm <agent> <globs...> --prompt "<task>" [--max-depth N] [--allowedTools TOOLS]
+```
+
+## Examples
+
+```bash
+coderlm codex "src/**/*.ts" --prompt "Find all TODO comments"
+coderlm codex "src/**" "lib/**" --prompt "Architecture overview"
+coderlm "bunx --bun @google/gemini-cli" "**/*.py" --prompt "Review for security issues"
+coderlm "bunx --bun @google/gemini-cli" "src/**" --prompt "Find dead code"
+coderlm claude "src/**" --prompt "Fix type errors" --allowedTools "Bash,Edit"
+```

coderlm-0.1.1/biome.jsonc

@@ -0,0 +1,4 @@
+{
+  "$schema": "./node_modules/@biomejs/biome/configuration_schema.json",
+  "extends": ["ultracite/biome/core"]
+}

coderlm-0.1.1/bun.lock

@@ -0,0 +1,70 @@
+{
+  "lockfileVersion": 1,
+  "configVersion": 1,
+  "workspaces": {
+    "": {
+      "name": "coding-agent-rlm",
+      "devDependencies": {
+        "@biomejs/biome": "2.4.0",
+        "ultracite": "7.2.3",
+      },
+    },
+  },
+  "packages": {
+    "@biomejs/biome": ["@biomejs/biome@2.4.0", "", { "optionalDependencies": { "@biomejs/cli-darwin-arm64": "2.4.0", "@biomejs/cli-darwin-x64": "2.4.0", "@biomejs/cli-linux-arm64": "2.4.0", "@biomejs/cli-linux-arm64-musl": "2.4.0", "@biomejs/cli-linux-x64": "2.4.0", "@biomejs/cli-linux-x64-musl": "2.4.0", "@biomejs/cli-win32-arm64": "2.4.0", "@biomejs/cli-win32-x64": "2.4.0" }, "bin": { "biome": "bin/biome" } }, "sha512-iluT61cORUDIC5i/y42ljyQraCemmmcgbMLLCnYO+yh+2hjTmcMFcwY8G0zTzWCsPb3t3AyKc+0t/VuhPZULUg=="],
+
+    "@biomejs/cli-darwin-arm64": ["@biomejs/cli-darwin-arm64@2.4.0", "", { "os": "darwin", "cpu": "arm64" }, "sha512-L+YpOtPSuU0etomfvFTPWRsa7+8ejaJL3yaROEoT/96HDJbR6OsvZQk0C8JUYou+XFdP+JcGxqZknkp4n934RA=="],
+
+    "@biomejs/cli-darwin-x64": ["@biomejs/cli-darwin-x64@2.4.0", "", { "os": "darwin", "cpu": "x64" }, "sha512-Aq+S7ffpb5ynTyLgtnEjG+W6xuTd2F7FdC7J6ShpvRhZwJhjzwITGF9vrqoOnw0sv1XWkt2Q1Rpg+hleg/Xg7Q=="],
+
+    "@biomejs/cli-linux-arm64": ["@biomejs/cli-linux-arm64@2.4.0", "", { "os": "linux", "cpu": "arm64" }, "sha512-u2p54IhvNAWB+h7+rxCZe3reNfQYFK+ppDw+q0yegrGclFYnDPZAntv/PqgUacpC3uxTeuWFgWW7RFe3lHuxOA=="],
+
+    "@biomejs/cli-linux-arm64-musl": ["@biomejs/cli-linux-arm64-musl@2.4.0", "", { "os": "linux", "cpu": "arm64" }, "sha512-1rhDUq8sf7xX3tg7vbnU3WVfanKCKi40OXc4VleBMzRStmQHdeBY46aFP6VdwEomcVjyNiu+Zcr3LZtAdrZrjQ=="],
+
+    "@biomejs/cli-linux-x64": ["@biomejs/cli-linux-x64@2.4.0", "", { "os": "linux", "cpu": "x64" }, "sha512-WVFOhsnzhrbMGOSIcB9yFdRV2oG2KkRRhIZiunI9gJqSU3ax9ErdnTxRfJUxZUI9NbzVxC60OCXNcu+mXfF/Tw=="],
+
+    "@biomejs/cli-linux-x64-musl": ["@biomejs/cli-linux-x64-musl@2.4.0", "", { "os": "linux", "cpu": "x64" }, "sha512-Omo0xhl63z47X+CrE5viEWKJhejJyndl577VoXg763U/aoATrK3r5+8DPh02GokWPeODX1Hek00OtjjooGan9w=="],
+
+    "@biomejs/cli-win32-arm64": ["@biomejs/cli-win32-arm64@2.4.0", "", { "os": "win32", "cpu": "arm64" }, "sha512-aqRwW0LJLV1v1NzyLvLWQhdLmDSAV1vUh+OBdYJaa8f28XBn5BZavo+WTfqgEzALZxlNfBmu6NGO6Al3MbCULw=="],
+
+    "@biomejs/cli-win32-x64": ["@biomejs/cli-win32-x64@2.4.0", "", { "os": "win32", "cpu": "x64" }, "sha512-g47s+V+OqsGxbSZN3lpav6WYOk0PIc3aCBAq+p6dwSynL3K5MA6Cg6nkzDOlu28GEHwbakW+BllzHCJCxnfK5Q=="],
+
+    "@clack/core": ["@clack/core@1.0.1", "", { "dependencies": { "picocolors": "^1.0.0", "sisteransi": "^1.0.5" } }, "sha512-WKeyK3NOBwDOzagPR5H08rFk9D/WuN705yEbuZvKqlkmoLM2woKtXb10OO2k1NoSU4SFG947i2/SCYh+2u5e4g=="],
+
+    "@clack/prompts": ["@clack/prompts@1.0.1", "", { "dependencies": { "@clack/core": "1.0.1", "picocolors": "^1.0.0", "sisteransi": "^1.0.5" } }, "sha512-/42G73JkuYdyWZ6m8d/CJtBrGl1Hegyc7Fy78m5Ob+jF85TOUmLR5XLce/U3LxYAw0kJ8CT5aI99RIvPHcGp/Q=="],
+
+    "balanced-match": ["balanced-match@4.0.3", "", {}, "sha512-1pHv8LX9CpKut1Zp4EXey7Z8OfH11ONNH6Dhi2WDUt31VVZFXZzKwXcysBgqSumFCmR+0dqjMK5v5JiFHzi0+g=="],
+
+    "brace-expansion": ["brace-expansion@5.0.2", "", { "dependencies": { "balanced-match": "^4.0.2" } }, "sha512-Pdk8c9poy+YhOgVWw1JNN22/HcivgKWwpxKq04M/jTmHyCZn12WPJebZxdjSa5TmBqISrUSgNYU3eRORljfCCw=="],
+
+    "citty": ["citty@0.2.1", "", {}, "sha512-kEV95lFBhQgtogAPlQfJJ0WGVSokvLr/UEoFPiKKOXF7pl98HfUVUD0ejsuTCld/9xH9vogSywZ5KqHzXrZpqg=="],
+
+    "commander": ["commander@14.0.3", "", {}, "sha512-H+y0Jo/T1RZ9qPP4Eh1pkcQcLRglraJaSLoyOtHxu6AapkjWVCy2Sit1QQ4x3Dng8qDlSsZEet7g5Pq06MvTgw=="],
+
+    "deepmerge": ["deepmerge@4.3.1", "", {}, "sha512-3sUqbMEc77XqpdNO7FRyRog+eW3ph+GYCbj+rK+uYyRMuwsVy0rMiVtPn+QJlKFvWP/1PYpapqYn0Me2knFn+A=="],
+
+    "glob": ["glob@13.0.5", "", { "dependencies": { "minimatch": "^10.2.1", "minipass": "^7.1.2", "path-scurry": "^2.0.0" } }, "sha512-BzXxZg24Ibra1pbQ/zE7Kys4Ua1ks7Bn6pKLkVPZ9FZe4JQS6/Q7ef3LG1H+k7lUf5l4T3PLSyYyYJVYUvfgTw=="],
+
+    "jsonc-parser": ["jsonc-parser@3.3.1", "", {}, "sha512-HUgH65KyejrUFPvHFPbqOY0rsFip3Bo5wb4ngvdi1EpCYWUQDC5V+Y7mZws+DLkr4M//zQJoanu1SP+87Dv1oQ=="],
+
+    "lru-cache": ["lru-cache@11.2.6", "", {}, "sha512-ESL2CrkS/2wTPfuend7Zhkzo2u0daGJ/A2VucJOgQ/C48S/zB8MMeMHSGKYpXhIjbPxfuezITkaBH1wqv00DDQ=="],
+
+    "minimatch": ["minimatch@10.2.1", "", { "dependencies": { "brace-expansion": "^5.0.2" } }, "sha512-MClCe8IL5nRRmawL6ib/eT4oLyeKMGCghibcDWK+J0hh0Q8kqSdia6BvbRMVk6mPa6WqUa5uR2oxt6C5jd533A=="],
+
+    "minipass": ["minipass@7.1.2", "", {}, "sha512-qOOzS1cBTWYF4BH8fVePDBOO9iptMnGUEZwNc/cMWnTV2nVLZ7VoNWEPHkYczZA0pdoA7dl6e7FL659nX9S2aw=="],
+
+    "nypm": ["nypm@0.6.5", "", { "dependencies": { "citty": "^0.2.0", "pathe": "^2.0.3", "tinyexec": "^1.0.2" }, "bin": { "nypm": "dist/cli.mjs" } }, "sha512-K6AJy1GMVyfyMXRVB88700BJqNUkByijGJM8kEHpLdcAt+vSQAVfkWWHYzuRXHSY6xA2sNc5RjTj0p9rE2izVQ=="],
+
+    "path-scurry": ["path-scurry@2.0.1", "", { "dependencies": { "lru-cache": "^11.0.0", "minipass": "^7.1.2" } }, "sha512-oWyT4gICAu+kaA7QWk/jvCHWarMKNs6pXOGWKDTr7cw4IGcUbW+PeTfbaQiLGheFRpjo6O9J0PmyMfQPjH71oA=="],
+
+    "pathe": ["pathe@2.0.3", "", {}, "sha512-WUjGcAqP1gQacoQe+OBJsFA7Ld4DyXuUIjZ5cc75cLHvJ7dtNsTugphxIADwspS+AraAUePCKrSVtPLFj/F88w=="],
+
+    "picocolors": ["picocolors@1.1.1", "", {}, "sha512-xceH2snhtb5M9liqDsmEw56le376mTZkEX/jEb/RxNFyegNul7eNslCXP9FDj/Lcu0X8KEyMceP2ntpaHrDEVA=="],
+
+    "sisteransi": ["sisteransi@1.0.5", "", {}, "sha512-bLGGlR1QxBcynn2d5YmDX4MGjlZvy2MRBDRNHLJ8VI6l6+9FUiyTFNJ0IveOSP0bcXgVDPRcfGqA0pjaqUpfVg=="],
+
+    "tinyexec": ["tinyexec@1.0.2", "", {}, "sha512-W/KYk+NFhkmsYpuHq5JykngiOCnxeVL8v8dFnqxSD8qEEdRfXk1SDM6JzNqcERbcGYj9tMrDQBYV9cjgnunFIg=="],
+
+    "ultracite": ["ultracite@7.2.3", "", { "dependencies": { "@clack/prompts": "^1.0.1", "commander": "^14.0.3", "deepmerge": "^4.3.1", "glob": "^13.0.3", "jsonc-parser": "^3.3.1", "nypm": "^0.6.5" }, "peerDependencies": { "oxlint": "^1.0.0" }, "optionalPeers": ["oxlint"], "bin": { "ultracite": "dist/index.js" } }, "sha512-WKNS2sKAZe4BHu+JGbZebXvy/A1QagDaBnndrK/zwOJAze/mQ8jeHfdG2bPlv3qcJ5fdS3w2Kd7c/eIcH78HvA=="],
+  }
+}

coderlm-0.1.1/justfile

@@ -0,0 +1,67 @@
+set shell := ["bash", "-euo", "pipefail", "-c"]
+
+# Show available recipes
+default:
+    @just --list
+
+# Run tests
+test:
+    bun test
+
+# Format code (py, ts, or all)
+fmt target="all":
+    just _fmt-{{target}}
+
+_fmt-all:
+    just _fmt-ts
+    just _fmt-py
+
+_fmt-ts:
+    bunx --bun ultracite fix
+
+_fmt-py:
+    ruff format
+    ruff check --fix
+
+# Lint code (py, ts, or all)
+lint target="all":
+    just _lint-{{target}}
+
+_lint-all:
+    just _lint-ts
+    just _lint-py
+
+_lint-ts:
+    bunx --bun ultracite check
+
+_lint-py:
+    ruff format --check
+    ruff check
+
+# Publish current version to npm, pypi, or both (default: all)
+publish target="all":
+    just _publish-{{target}}
+
+# Bump version in package.json and pyproject.toml (default: patch)
+bump level="patch":
+    just _bump {{level}}
+
+# Internal bump implementation
+_bump level:
+    #!/usr/bin/env bash
+    set -euo pipefail
+    npm version {{level}} --no-git-tag-version
+    version=$(jq -r .version package.json)
+    perl -pi -e "s/^version = .*/version = \"$version\"/" pyproject.toml
+    echo "Bumped to $version"
+
+_publish-all:
+    just _publish-npm
+    just _publish-pypi
+
+_publish-npm:
+    npm publish
+
+_publish-pypi:
+    uv build
+    uv publish

coderlm-0.1.1/package.json

@@ -0,0 +1,37 @@
+{
+  "name": "coderlm",
+  "version": "0.1.1",
+  "description": "Process large codebases using the RLM (Recursive Language Model) pattern",
+  "bin": {
+    "coderlm": "src/coderlm"
+  },
+  "files": [
+    "src/coderlm",
+    "src/bashrlm.sh",
+    "src/bashrlm.md",
+    "SKILL.md",
+    "README.md"
+  ],
+  "keywords": [
+    "cli",
+    "rlm",
+    "claude",
+    "codex",
+    "agent",
+    "llm"
+  ],
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/zenbase-ai/coderlm.git"
+  },
+  "devDependencies": {
+    "@biomejs/biome": "2.4.0",
+    "ultracite": "7.2.3"
+  },
+  "scripts": {
+    "test": "bun test",
+    "check": "ultracite check",
+    "fix": "ultracite fix"
+  }
+}

coderlm-0.1.1/pyproject.toml

@@ -0,0 +1,22 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "coderlm"
+version = "0.1.1"
+description = "Process large codebases using the RLM (Recursive Language Model) pattern"
+readme = "README.md"
+license = "MIT"
+requires-python = ">=3.9"
+
+[project.scripts]
+coderlm = "src:main"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src"]
+
+[dependency-groups]
+dev = [
+    "ruff>=0.15.1",
+]