New advertisers: 50% off your first month with code FIRSTMONTHSee slots

Maps a toxic family system into persistent context, one linked file per person, for therapy intake.

version
1.0.1
GitHub stars
32
View on GitHub

What it does

  • Adds capture-family-system, analyse-dynamics, map-family-system, and prepare-intake-pack skills for mapping and printing a family's harm patterns
  • Covers roles, triangles, DARVO, and parentification using Bowen, Minuchin, and related family-systems research, with evidence tiers
  • Keeps your words and real names private in a git repo, with aliases applied to everything shown to you
  • Designed for people in harmful family systems who want an intake pack to bring to a therapist

Map a toxic family system: one file per person, roles grounded in family-systems research, printable relationship diagram, therapy intake pack, aliases for names that hurt, private git repo plus optional vector store.

Add marketplace

/plugin marketplace add danielrosehill/Claude-Code-Plugins

Run it inside Claude Code once, then install the plugin below.

Install

/plugin install family-systems-context@danielrosehill

Run it inside Claude Code after adding the marketplace.

Files

Discovered from source

Analyse the dynamics

Take the recorded system apart through the research lenses in docs/frameworks.md: who originates harm, who carries it, who is protected, who absorbs it, which structures hold it in place, and what repeats across generations. The output is written into the overview and actor files, so it serves as reference for therapy and for other tools.

This is analysis of a system, not therapy and not a verdict on the user. The user's account is the material; the agent does not test it.

Load

Work from the workspace root: every command below runs there and uses the workspace's own scripts/.

  1. Run ../../scripts/workspace.sh relative to this skill's directory.
  2. Read profile.env and analyse-dynamics.md in the overlay if present.
  3. Read AGENTS.md, docs/frameworks.md in full, docs/glossary.md, vocabulary.md, family-overview.md and every actor file. Read sources only where an event needs checking for detail.

Lenses, in order

For each, look for evidence in the recorded events; name the concept and tier; link the actors and events. Skip a lens with nothing to say rather than forcing it.

  1. Layers of harm. From the centre outward: originators (whose behaviour drives the harm), carriers (who act for them: recruited relatives, "flying monkeys"), protected members (favoured, "golden child"), bystanders and enablers (who see and do nothing, or cover), absorbers (scapegoat, parentified child, the user). One member can sit in more than one layer, and positions can change over time.
  2. Structure (Minuchin): subsystems, boundaries (enmeshed, disengaged, clear), hierarchy and where it is inverted, coalitions and cross-generational coalitions, triangulation, detouring.
  3. Triangles and process (Bowen): the main triangles and how tension moves around them; family projection onto a child; over- and under-functioning; cutoffs between other members; multigenerational transmission.
  4. Roles with their functions; the user's terms with their mappings (Tier C role labels recorded as such).
  5. Parentification: instrumental or emotional, by whom, from when.
  6. Abuse-specific patterns: DARVO sequences, gaslighting, coercive-control patterns, ostracism or silent treatment, hoovering cycles. Quote the recorded events.
  7. Loyalties and ledger (contextual): invisible loyalties, split loyalty, destructive entitlement passed down.
  8. Functioning: a short description against each Circumplex and McMaster dimension.
  9. Family rules: what keeps the system closed.

Write

  • Overview: Layers of harm, Structure, Functioning, Recurring dynamics, Multigenerational patterns, Family rules. Each dynamic: one line naming it, the lens and tier, the actors, and the events it rests on (linked).
  • Actor files: add the lenses that apply under Interpretations and lenses, labelled as the agent's analysis with lens and tier; update Roles and roles: frontmatter.
  • Relationships: add any coalitions, recruitments, favouring or scapegoating edges the analysis surfaced to the overview's relationships list.
  • Gaps: where a lens would apply but events are thin, add an Open questions item ("Were there other times the aunt raised things on mother's behalf?"), not a hedge.
  • History and Change log lines: ": dynamics analysis; added ".

Then run uv run --script scripts/family_map.py all and commit: git commit -m "Analyse dynamics <date>".

Report to the user

A short spoken-style summary: the layers, the two or three central dynamics, and what is still unknown. Offer map-family-system and prepare-intake-pack. Do not offer advice on what to do about the family unless asked.

Data storage

Reads profile.env and analyse-dynamics.md from the overlay folder. Writes only inside the workspace and commits.

Readme

Family Systems Context

An agent plugin for mapping a toxic or abusive family system and keeping that map as persistent context: one linked markdown file per person, a family overview, roles and dynamics grounded in family-systems research, a printable relationship diagram, a PDF intake pack for a therapist, aliases for names that are hard to see, all in a private git repository that is the source of truth, with an optional vector store mirrored from it.

What it is, and isn't

Some people find themselves in family systems they did not choose and that harm them: many people, shifting alliances, roles passed between siblings, tactics that only show as a pattern over years. This plugin uses what agents are good at, holding a lot of detail and its relationships, to break such a system down and plot it: who originates harm, who carries it ("flying monkeys"), who is protected, who absorbs it.

It is a sidecar to therapy, not therapy. The agent captures, structures and maps; a therapist does the therapeutic work, with this as intake or reference material. Other AI tools can read it too.

  • The user is believed. Their account is the record.
  • The user's words are kept. Terms from the informal abuse-recovery literature are used as the user uses them, and mapped to research constructs in a glossary a clinician can read.
  • Research-grounded. Bowen family systems theory, Minuchin's structural family therapy, genogram conventions, contextual therapy, parentification research, DARVO, betrayal trauma, coercive control, ACEs, and the Circumplex and McMaster family-functioning models, each with an evidence tier.
  • Names that hurt. The record keeps real names; aliases are applied to everything shown to the user, and pseudonymised or obfuscated copies can be made for others.

How it is organised

The plugin is public and generic. Your family data lives in a workspace: a private repository created from the public Family-System-Context-Template, which carries its own skills and docs and works with this plugin uninstalled. Your settings live in an overlay folder outside the plugin (profile.env).

family-overview.md     the system as a whole; the relationship list that draws the diagram
actors/                one file per person, including self.md for you
sources/               raw material, verbatim (voice-note transcripts, imported notes)
vocabulary.md          your own terms and what you mean by them
aliases.yaml           comfort aliases and share labels
docs/                  method, frameworks, glossary, diagram, aliases, vector store

Skills

Setting up (plugin only)

  • setup: settings: where the workspace is, pushing, voice-note transcription, vector store.
  • create-workspace: a new private repository from the template, a local folder, or an existing family-notes repository adopted in place.

Getting the family in

  • capture-family-system: guided interview, one question at a time, written as it goes.
  • ingest-voice-note: one long voice note (audio or transcript) in; transcript kept verbatim; changes extracted and applied after you see them.
  • import-existing-notes: notes you already have, mapped in; originals untouched.

Keeping it current and using it

  • update-family-system: incidents, contact changes, role shifts, corrections, with dated, append-only history.
  • analyse-dynamics: the research lenses: layers of harm, structure, triangles, roles, parentification, DARVO and gaslighting, loyalties, functioning.
  • map-family-system: the relationship diagram as a printable PDF and a Mermaid map.
  • prepare-intake-pack: a PDF for a therapist: diagram, overview, terms mapped to research, a section per person.
  • load-family-context: the right context at the start of a conversation.
  • manage-aliases: comfort aliases and share labels, with the printed key.
  • share-safely: aliased, pseudonymised or obfuscated copies for less trusted readers.
  • sync-vector-store: idempotent upsert into Qdrant (server or local folder) with any OpenAI-compatible embeddings endpoint, including local Ollama.

Installation

/plugin marketplace add danielrosehill/Claude-Code-Plugins
/plugin install family-systems-context@danielrosehill

Then run the setup skill, then create-workspace.

Requirements: git, gh (for a GitHub-backed workspace), uv for the scripts, Graphviz for the diagram PDF, pandoc (3.1.2+) and typst for the intake pack. Vector store: Qdrant, or nothing extra for its embedded local mode, plus an embeddings endpoint.

Settings

All optional; documented in reference/profile-keys.md. The overlay folder resolves, first match wins: $FAMSYS_HOME, $CLAUDE_USER_DATA/family-systems-context/, ~/.claude-user-data/family-systems-context/ if that root exists, ${XDG_DATA_HOME:-~/.local/share}/claude-plugins/family-systems-context/ if that root exists, else ~/.claude-user-data/family-systems-context/. scripts/workspace.sh prints which.

Maintaining

The template repository is canonical for everything under template/, the workspace skills and the scripts. Change them there, then run scripts/sync-from-template.sh; --check reports drift. Plugin-owned: setup, create-workspace, reference/, the sync script.

Not a crisis service

Every workspace has a SUPPORT.md. If you are at risk, contact a person: https://findahelpline.com.

License

MIT

Details

Marketplace
danielrosehill/Claude-Code-Plugins
Source
danielrosehill/Family-Systems-Context-Plugin
Author
Daniel Rosehill
Category
Health
License
MIT

More from danielrosehill/Claude-Code-Plugins

All 264

Similar to family-systems-context

More Memory & Context