# Rcr

> Use when the user mentions rcr, Red Cliff Record, personal knowledge base, records, links, graph-style queries, or sync/integration runs for that KB—even if they omit the tool name. Runs and interprets the `rcr` CLI.

- Skill: `aias/rcr` (Agent Skill)
- Install (CLI): `npx skillmds@latest add aias/rcr`
- Raw SKILL.md: https://api.skillmd.com/api/skills/aias/rcr/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: Aias (https://skillmd.com/u/aias)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/aias/rcr

---


# Red Cliff Record CLI

Directory name is `red-cliff-record`; frontmatter `name` remains `rcr` — cross-link as `/rcr`.

The `rcr` CLI provides direct access to Red Cliff Record, a highly networked, manually curated knowledge base.

## Source Of Truth

- Trust live CLI help and errors over this document if they diverge.
- The command surface is implemented in `src/server/cli/rcr/`.

## Live CLI Help

Run `rcr --help` before composing a command; the flags below are a summary, and the installed CLI is the source of truth. Report the CLI as unavailable if the command fails rather than guessing its surface.

## Safety Defaults

- Prefer read commands first (`list`, `get`, `search`) before mutations.
- Ask for confirmation before destructive operations:
  - `rcr records delete ...`
  - `rcr records merge ...`
  - `rcr links delete ...`
  - `rcr db restore ...`
  - `rcr db reset ...`
  - `rcr db clone-prod-to-dev ...`
- Use `--dry-run` for destructive-capable `db` operations when available.
- Add `--limit` to `list`/`search` commands unless the user explicitly wants broad output.

## Global Behavior

- Default output: JSON wrapper

```json
{"data":<result>,"meta":{"count":N,"duration":M}}
```

- Error output (stderr, exit code 1):

```json
{ "error": { "code": "ERROR_CODE", "message": "..." } }
```

- `--raw` outputs just `data` without `{data,meta}` wrapper.
- `--format=table` prints human-readable tables.
- Global options:
  - `--format=json|table`
  - `--raw`
  - `--limit=N`
  - `--offset=N`
  - `--dev` (uses development DB connection)
  - `--debug`
  - `--help`, `-h`
  - `--version`, `-v`
- Practical note: `--limit` and `--offset` are most useful on list/search commands. Some strict subcommands reject them.
- Boolean flags support all forms:
  - `--flag`
  - `--flag=true`
  - `--flag=false`
- Command alias: `record` is accepted as `records`.

## Data Model

- Records: entities, concepts, artifacts
- Links: typed edges between records
- Media: files/URLs attached to records (image, video, text, etc.)
- Sources: integration provenance on records

## Command Reference

### `records`

- `rcr records get <id...> [--links]`
  - Single ID returns one record or NOT_FOUND error
  - Multi-ID returns array; missing records return `{ id, error: "NOT_FOUND" }`
- `rcr records list [filters]`
  - Filters:
    - `--type=entity,concept,artifact`
    - `--source=<comma-separated integration types>`
      - `ai_chat, airtable, browser_history, crawler, embeddings, feedbin, github, lightroom, manual, raindrop, readwise, twitter`
    - `--has-title[=BOOL]`
    - `--curated[=BOOL]`
    - `--private[=BOOL]`
    - `--parent[=BOOL]`
    - `--media[=BOOL]`
    - `--embedding[=BOOL]`
    - `--rating-min=N` and `--rating-max=N` (0-3)
    - `--order=field:dir,...`
      - Fields: `recordUpdatedAt, recordCreatedAt, title, contentCreatedAt, contentUpdatedAt, rating, id`
      - Direction: `asc|desc`
    - `--full` to fetch complete records instead of ID-only rows
- `rcr records create '<json>'`
- `rcr records update <id> '<json>'`
- `rcr records bulk-update <id,id,...> '<json>'`
- `rcr records delete <id...>`
- `rcr records merge <sourceId> <targetId>`
- `rcr records embed <id...>`
- `rcr records tree <id...>`
- `rcr records children <id>`
- `rcr records parent <id>`

### `media`

- `rcr media get <id...> [--with-record]`
- `rcr media list [filters]`
  - `--type=application|audio|font|image|message|model|multipart|text|video`
  - `--alt-text[=BOOL]`
  - `--record=<recordId>`
  - `--order=recordCreatedAt|recordUpdatedAt|id`
  - `--direction=asc|desc`
- `rcr media create --record <id> --file <path> [--name <filename>] [--type <mime>]`
- `rcr media create --record <id> --url <url>`
  - Exactly one of `--file` or `--url` is required
- `rcr media update <id> '{"altText":"..."}'`
- `rcr media generate-alt <id...> [--force]`

### `search`

- Hybrid (default): `rcr search <query> [--limit=N]`
- Text: `rcr search text <query> [--type=entity|concept|artifact] [--limit=N]`
- Semantic: `rcr search semantic <query> [--limit=N] [--exclude=id,id,...]`
- Similar by record: `rcr search similar <id...> [--limit=N]`

### `links`

- `rcr links list <recordId> [--predicate=<slug>] [--direction=incoming|outgoing]`
- `rcr links create '<json>'`
  - JSON shape: `{ sourceId, targetId, predicateId, notes? }`
- `rcr links delete <id...>`
- `rcr links predicates`

### `browsing`

- `rcr browsing daily <YYYY-MM-DD>`
- `rcr browsing omit`
- `rcr browsing omit-add <pattern>` (SQL `LIKE` syntax, `%` wildcard)
- `rcr browsing omit-delete <pattern...>`

### `github`

- `rcr github daily <YYYY-MM-DD>`
- `rcr github get <commitId>`

### `sync`

- `rcr sync`
  - Runs daily sync set, then enrichments:
    - Daily integrations: `browsing, raindrop, readwise, github, airtable, twitter`
    - Then: avatars, alt-text, embeddings enrichments
- `rcr sync <integration>`
  - Valid integrations:
    - `github, readwise, raindrop, airtable, adobe, feedbin, browsing, twitter, agents`
  - Single integration run is still followed by enrichments.

### `enrich`

- `rcr enrich`
  - Runs all: `avatars -> alt-text -> embeddings`
- `rcr enrich avatars`
- `rcr enrich alt-text [--limit=N]`
- `rcr enrich embeddings`
- `--limit` is only valid with `rcr enrich alt-text`.

### `db`

- `rcr db backup <prod|dev> [--data-only] [--dry-run|-n]`
- `rcr db restore <prod|dev> [--clean] [--data-only] [--file <path>|--file=<path>|-f <path>|-f=<path>] [--dry-run|-n] [--yes|-y]`
  - If `--file/-f` is omitted, restore auto-selects the newest matching backup for the target environment:
    - Data-only restore: `<target>-data-*.dump`
    - Full restore: `<target>-*.dump` excluding `<target>-data-*.dump`
  - `--dry-run` uses a placeholder dump path when no matching file exists.
- `rcr db reset [dev]` (dev only)
- `rcr db seed [dev]` (dev only)
- `rcr db status`
  - Connection determined by `--dev` flag / `NODE_ENV`. Use `--dev` for dev, omit for prod.
- `rcr db clone-prod-to-dev [--dry-run|-n] [--yes|-y]`

### `fetch`

- `rcr fetch url <url> [--content-only]`
- Uses Jina Reader (`https://r.jina.ai/`) to return parsed markdown.

## Practical Examples

```bash
# Get one record with links
rcr records get 123 --links

# Find uncurated entities/concepts
rcr records list --type=entity,concept --curated=false --limit=25

# Bulk curate
rcr records bulk-update 101,202,303 '{"isCurated":true}'

# Find media missing alt text
rcr media list --type=image --alt-text=false --limit=50

# Generate alt text for specific media items
rcr media generate-alt 123 456 789 --force

# Hybrid search (default search mode)
rcr search "recursive prompting patterns" --limit=10

# Semantic search excluding known IDs
rcr search semantic "knowledge graphs" --exclude=12,34 --limit=15

# Add an omit pattern to browsing summaries
rcr browsing omit-add "%localhost%"

# Run one integration sync + automatic enrichments
rcr sync github --debug

# Run only alt-text enrichment with a smaller batch
rcr enrich alt-text --limit=20

# Preview a prod->dev clone without changes
rcr db clone-prod-to-dev --dry-run

# Emit raw output for downstream tooling
rcr records list --source=readwise --raw | jq '.[].id'

# Fetch clean markdown from a URL
rcr fetch url https://example.com/article --content-only --raw
```