memories ingest

memories ingest [source] [options]

Scan existing IDE rule files and extract memories from them. This is the fastest way to migrate an existing setup to memories.sh. Supports both flat file ingestion and structured directory-based ingestion with full metadata extraction.

Arguments

Argument	Description
`source`	Source name or file path (see sources below)

Sources

Flat-file sources

Source	Scanned Path	Description
`cursor`	`.cursorrules`, `.cursor/rules/memories.mdc`	Cursor rules (legacy + current)
`claude`	`CLAUDE.md`	Claude Code instructions
`agents`	`AGENTS.md`	Multi-tool agents file
`copilot`	`.github/copilot-instructions.md`	GitHub Copilot
`windsurf`	`.windsurfrules`, `.windsurf/rules/memories.md`	Windsurf rules
`cline`	`.clinerules/memories.md`	Cline rules
`roo`	`.roo/rules/memories.md`	Roo rules
`gemini`	`GEMINI.md`	Gemini instructions

Directory-based sources

These sources parse structured configuration files and extract rich metadata (paths, category, skill metadata):

Source	Scanned Path	What's Extracted
`claude-rules`	`.claude/rules/*.md`	Path-scoped rules with `paths:` frontmatter
`cursor-rules`	`.cursor/rules/*.mdc`	Path-scoped rules with `globs:` frontmatter
`skills`	`.agents/skills/`, `.claude/skills/`, `.cursor/skills/`, `.codex/skills/`, `.kiro/skills/`, `.kilo/skills/`, `.trae/skills/`, `.antigravity/skills/`, `.goose/skills/`, `.windsurf/skills/`, `.cline/skills/`, `.clinerules/skills/`, `.roo/skills/`, `.amp/skills/`, `.opencode/skills/`, `.factory/skills/`, `.gemini/skills/`	Skill memories with metadata
`.agents/`	`.agents/` directory	Full canonical config (instructions, rules, skills)

Options

Option	Description
`--type <type>`	Override type for all imported memories
`--dry-run`	Preview without importing
`--all`	Scan all known locations (flat files + directories)
`--no-dedup`	Skip duplicate detection

Examples

Ingest from a specific tool:

memories ingest cursor
memories ingest claude

Ingest path-scoped rules from Claude Code:

memories ingest claude-rules

Ingest path-scoped rules from Cursor:

memories ingest cursor-rules

Ingest skills:

memories ingest skills

Ingest from the canonical .agents/ directory:

memories ingest .agents/

Ingest from all known locations:

memories ingest --all

Preview what would be imported:

memories ingest cursor --dry-run

Ingest from a custom file:

memories ingest ./my-rules.md --type rule

Directory-Based Ingestion

When ingesting from structured directories, memories.sh extracts richer metadata than flat-file ingestion:

Claude Rules (`.claude/rules/*.md`)

Parses YAML frontmatter for paths: globs
Sets the memory's paths field from the frontmatter
Derives category from the filename (e.g., api.md → category api)
Extracts bullet points as individual rule memories

Cursor Rules (`.cursor/rules/*.mdc`)

Parses MDC frontmatter for globs: patterns
Translates globs: → paths on the memory
Handles alwaysApply: true rules as non-path-scoped
Derives category from the filename

Skills (`/skills//SKILL.md`)

Parses YAML frontmatter for name, description, and other skill metadata
Stores frontmatter as the memory's metadata JSON field
Uses the parent directory name as the category
Ingests the full markdown body as memory content

Behavior

Extracts bullet points, numbered items, and significant plain text from rule files
Auto-infers memory type from content keywords:
- "always", "never", "must", "should" → rule
- "chose", "decided", "because" → decision
- Specific values, numbers, limits → fact
Skips files that were generated by memories.sh (detected by marker comment)
Deduplicates against existing memories using content and path-aware comparison
Scans both current and legacy file locations (e.g., .cursorrules, .windsurfrules)

On this page