ase

Reviewed at commit 18f93ac New version, not yet reviewed

Caution: Pre-approves Bash git and rm -rf in its merge, resolve and import skills, and a hook auto-approves ASE commands and, while lint skills run, file edits.

Sends data to Optional chat and search services you set up yourself with ase setup mcp (OpenAI, Gemini, Grok, DeepSeek, Qwen, GLM, Brave, Exa, Perplexity), only when you set a key

Agentic Software Engineering (ASE)

What we checked

A large toolkit of about 60 skills for planning, linting, reviewing, importing, staging and merging code, with a terminal layer that adds forms, manual pages, a live task view and inline images and diagrams above or in the transcript. It needs the separately installed ase npm tool, which its hooks and MCP server run on every session, prompt and tool call; diagrams and images are rendered locally by that tool.

Review details

Static review of the source at the pinned commit; not run. Full re-review of version 2.0.12 (the diff since f8d7d97 was too large for a delta review), looking closely at what changed. Read in full: plugin.json, hooks/hooks.json (unchanged), mods/ase-mods.ts, the diff of mods/ase-mods-common.ts, and the new mods/ase-mods-task-edit.ts and mods/ase-mods-render-image.ts; the other mods files (only renamed, unchanged) were checked by targeted search, which confirms that ase-mods.ts is the only module touching the engine. The module registers session.start, tool.check, tool.call, prompt.edit, prompt.submit, ui.scroll, ui.render (AssistantMessage, AbovePrompt, ToolUse, ToolResult) and command.describe; it reads files under the plugin root with $.fs.read and $.fs.list, fills and reads the prompt with $.prompt, polls the prompt every 100 ms while a slash command is typed, and for the intent help (typing !!) sends the session transcript to the model through $.model.fork or $.model.complete (uses tokens on your account). New in this version: in an interactive terminal session.start reads TERM, TERM_PROGRAM, TMUX, STY, KITTY_WINDOW_ID and GHOSTTY_RESOURCES_DIR with $.env.get, runs tmux display -p through $.process.run to probe for inline image support, and registers three tools with $.tool.register: ase_task_band (shows a task in a band above the prompt and reloads it every 2 s through the ase MCP tool ase_task_load via $.tool.call, with a word diff of the last change), ase_mods_render_image (an SVG, PNG, GIF or JPEG given as file path, relative or absolute, or as data) and ase_mods_diagram (Mermaid or D2). Its tool.check hooks answer allow for these three tools without a prompt; the image tool passes the path to the local ase MCP tool ase_rasterize, reads the resulting temporary PNG with $.fs.read and draws it with ui.Image and $.ui.blit, while the model only gets OK or ERROR text, not the image. The targeted check of the ase CLI source for rasterizing and diagrams found local rendering only (resvg-wasm, beautiful-mermaid, @d2lang/d2 wasm, temp files under the system temp folder), no network. The options.enabled setting turns the module off. hooks.json runs shell hooks calling the ase CLI (SessionStart, SessionEnd, PreToolUse, UserPromptSubmit, Stop), and plugin.json starts an MCP server with ase mcp, which spawns the ase service in the background on a local port. The CLI is not in the plugin folder (README: npm install -g @rse/ase, then ase setup install); its hook source was checked by diff: the session-start hook now also writes alias shim skills (user-invocable only, forwarding to an ASE skill) into ~/.claude/skills and the project's .claude/skills when aliases are configured with the new ase-meta-alias skill or a project .ase/aliases.yaml, and removes its own orphaned shims; task store and relay clients gained an opt-in insecure TLS mode for self-configured servers. Agents, task boards, relay server and telemetry setup were not reviewed. As before, the PreToolUse hook answers allow for Bash commands starting with ase without shell metacharacters, ASE skills and MCP tools, the optional foreign chat and search MCP tools, Read below the plugin root, and Edit while the lint and docs skills run. Skills pre-approve Bash(git *), Bash(rm -rf *), Bash(rm -f *), Write and Edit in merge, resolve, dissect, stage and changelog skills; the new ase-repo-import skill pre-approves Bash(git *), mkdir -p, cp -p, rm -rf, Read, Edit and Write, fetches a pull or merge request or commit from a remote or a forge URL you give (git fetch), applies it to the working copy with git apply --3way or --reject, writes conflict markers, unstages with git reset -- on the merged paths and removes its temp patch in .git with rm -rf. The new ase-meta-alias skill pre-approves Bash(ase alias *). Other skill changes: a --focus option and an --interactive task band for ase-task-edit and ase-task-grill, ase-task-dissect re-wiring successor tasks through the ase MCP task tools, and diagrams routed to the inline diagram tool with the ase-meta-diagram agent as fallback. ase-arch-discover still pre-approves curl -s for Maven search. No unpinned npx in skills. Skill Markdown was read through the diff and by targeted search; the roughly 1.1 MB of prompt prose was not read in full.

What it uses
  • Reads your prompts: yes
  • Sees tool calls: yes
  • Blocks or changes tool calls: no
  • Reads or writes files: yes
  • Runs programs on your machine: yes
  • Calls the model (uses tokens): yes
  • Sends data over the network: yes
  • Stores data: yes
  • Shows something on screen: yes
Hooked events
Screenshot of ase
Made this mod? Add a badge to your README

modsforClaude badge for ase

Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.

[![Listed on modsforClaude](https://modsforclaude.com/badge/ase.svg)](https://modsforclaude.com/mods/ase/)

← All mods