archcore
Reviewed
at commit 3f370e6
New version, not yet reviewed
Caution: Offers during /archcore:init to install the separate archcore CLI by piping archcore.ai's install script into bash; its hooks and document server then run that unreviewed program.
Spec-driven development and context engineering for AI coding agents, backed by project context in Git.
0 upvotes
Log in to upvote or save- Made by
- Archcore
- Pinned commit
-
3f370e6bb04e49658cdc6858961c833bf28ba414 - What we checked
-
Adds a band above the prompt that suggests one Archcore command after a turn (record a decision, review a pushed branch) and starts the separate archcore program on your machine as an MCP server and as session and write-guard hooks for the project's .archcore/ documents. Needs the archcore CLI installed by you. It blocks edits that bypass its document tools; /archcore:init can wire project files after you type confirm.
Review details
Static review of the source at the pinned commit; not run. Delta review of the changes since e47d597; the rest carries over from the review at that commit. The TSX module (unchanged) registers session.start, session.end, prompt.submit, skill.prompt, turn.start, turn.complete, tool.call (Write, Edit, NotebookEdit, ExitPlanMode, Agent Plan, Bash matching git/gh/glab, and Archcore MCP tools), prompt.suggest and ui.render AbovePrompt. It keeps counters in plugin state, classifies prompts with keyword regexes, calls only the Archcore MCP tools search_documents and get_document through $.mcp.call, reads a plan file inside the project with $.fs.read, and offers the command as the Tab suggestion. It makes no network calls and spawns nothing itself; the next_step_hints option turns the module off. hooks.json also declares three shell hooks: bin/session-start, bin/pre-tool-use (matcher Write|Edit) and bin/post-tool-use (Archcore document tools). They are POSIX scripts that skip plugin-cache directories, fail open when the CLI is missing or older than 0.7.0 and pipe the raw hook payload (which for pre-tool-use includes the file path and new content) into the archcore CLI (archcore hooks claude-code ...); the write guard over .archcore/ documents lives in that binary. session-start also runs archcore update --check at most once a day; that is a version check by the CLI, and its host was not reviewed. The MCP server is the command archcore mcp from PATH. If the CLI is missing, session-start tells Claude to tell you the install one-liner (curl | bash, irm | iex); nothing installs it. Changes in this version (0.15.2 to 0.15.5): version number in the manifests and agent files, and the list_relations description in the agents notes that an unfiltered listing stops at 300 edges; no code, hook, command or install changes. Earlier delta carried over (0.14.3 to 0.15.2): bin/session-start no longer tells Claude to initialize .archcore/ in a folder without one; it asks which project to use (and, with CLI 0.15.0+, defers to the CLI's project listing); the skills gained a shared target-project.md so tool calls can pass a project argument for other projects found under paths in the request, and global sources are now treated as writable in their source project when the user asks (the skills require asking if local or source is unclear, and naming the source in the report); a new hung-plan check in the actualize/review tracks flags shipped or abandoned plan documents and, after one batch confirmation, removes them with remove_document (git keeps the file; an uncommitted plan is named as unrecoverable first); describe track tightened. No new hosts, installs, allowed-tools or bypassPermissions. Skills and agents use the Archcore MCP tools, Read, Grep and Glob; no allowed-tools or unpinned package runs. The init skill can write .mcp.json, .claude/settings.json (a SessionStart hook), CLAUDE.md and AGENTS.md blocks, and seed .archcore/ documents, but its text requires a preview and the user typing confirm first. Not reviewed: the archcore binary (CLI, MCP server and what it does with payloads), npm dependencies (none), and the 58 prompt files were only searched for commands, installs, URLs and settings writes, not read in full.
- What it uses
-
- Reads your prompts: yes
- Sees tool calls: yes
- Blocks or changes tool calls: yes
- Reads or writes files: yes
- Runs programs on your machine: yes
- Calls the model (uses tokens): no
- Sends data over the network: no
- Stores data: yes
- Shows something on screen: yes
Made this mod? Add a badge to your README
Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.
[](https://modsforclaude.com/mods/archcore/)