CRUX-Compress: Command for Claude Code

crux-recall

Decompress, query, and display CRUX memories in human-readable form.

Repository: github.com/zotoio/CRUX-Compress

Usage

/crux-recall                                    - Show contextually relevant memories
/crux-recall "search query"                     - Search memories by keyword
/crux-recall 20260403-crux-memories              - Show memories from a specific spec
/crux-recall memories/core/some-memory.memory.md - Display a specific memory file
/crux-recall --total                             - Visualize the entire memory system as an interactive canvas

Parameters

Parameter

Description

--total

Generate an interactive canvas visualization of the entire memory system using only cursor/canvas SDK primitives. The canvas includes: summary stats, type/search filters, an SVG force-directed graph (2D, pre-computed at module scope via Verlet simulation), a detail panel for selected memories, type/strength distribution charts (PieChart/BarChart), and a filterable memory table. Nodes represent memories (sized by strength, colored by type, labeled by title). Edges connect memories sharing tags or source specs. Supports click-to-detail, hover highlighting, type filter toggles, and text search.

Instructions

When this command is invoked, spawn a crux-memory-recall subagent to query and display memories. Recall is read-only — it never modifies memory files on disk.

User input escalation — CRITICAL: This command uses Pattern B (work first, then escalate). The subagent NEVER calls AskQuestion directly — the parent agent handles every post-display prompt (delete, consolidate, promote, skip) and resumes the subagent with the answers. For the full Pattern A / Pattern B contract and the needs_user_input YAML schema, see .cursor/skills/_memory-shared.md#user-input-escalation and AGENTS.md.

Argument Handling

• No arguments: The subagent loads the memory index (.crux/memory-index.yml) and surfaces memories most likely to be relevant to the current conversation context. For each memory, it shows title, type, strength, reference count, and a brief rationale for why it was surfaced.
• Quoted text (e.g. "performance optimization"): The subagent searches existing memories by title, description, tags, and body content. Results are ranked by relevance with decompressed body content shown for compressed memories. Pass $ARGUMENTS to the subagent as the search query.
• Spec name(s) (e.g. 20260403-crux-memories): The subagent finds all memories whose source field matches the given spec slug(s). Results are grouped by type. Pass $ARGUMENTS to the subagent as the spec name(s).
• File path(s) (e.g. memories/learning/foo.memory.md): The subagent reads the specified memory file(s) directly. Compressed files (.memory.crux.md) are decompressed for display. Full frontmatter and body are shown. Pass $ARGUMENTS to the subagent as the file path(s).
• --total: The subagent gathers the complete memory corpus — index metadata, all memory files, and their relationships — then generates an interactive canvas visualization of the entire memory system using only cursor/canvas SDK components and inline SVG. The crux-memory-recall agent owns the structural template loaded on this cold path. No table or text output is produced; the canvas is the deliverable.