Psychology Companion is a professional, evidence-aware AI assistant for reflection, recurring patterns, relationships, and sexuality. It is designed for thoughtful conversation, not diagnosis or emergency care.
The plugin contains three independently available skills. psychologist is
the lightweight conversational core, psychology-journal provides optional
private continuity and Git workflows, and psychology-research provides
source-backed research. A focused sexology specialty remains part of the core
and loads only when the topic calls for it.
The root development evidence archive preserves the dated psychological and sexological research used to design these skills. It is available for maintainer and auditor review but is not loaded during normal plugin use. The archive documents the project's evidence basis and limits; it does not certify the plugin as a clinician, validated treatment, or diagnostic system.
Psychology Companion is an AI instruction plugin. It is not a psychologist, psychotherapist, physician, or other licensed or certified professional, and it does not create a clinician-patient relationship. Its output is for reflection and general information; it is not a diagnosis, treatment plan, medical advice, or emergency service.
Before acting on guidance that could affect mental or physical health, safety, relationships, or another major life decision, review it with an appropriately licensed or certified psychologist, psychotherapist, physician, or other qualified professional who can assess the full circumstances. Do not use this plugin as the sole basis for diagnosis, treatment, medication, or crisis decisions. In an immediate danger or medical emergency, contact verified local emergency or crisis services.
- Uses professional, empathic, collaborative communication adapted to the user, context, culture, and level of reactance.
- Marks whether a claim comes from a document, self-report, observation, or hypothesis, and leaves unknowns explicit.
- Uses structured methods only when they fit the user's goal.
- Treats sexual interests without automatic pathologizing while protecting consent, safety, and autonomy.
- When the optional journal skill is available and longitudinal mode is active, records substantive sessions and updates continuity files.
- Researches psychological questions on demand without loading the research workflow into ordinary supportive conversations. The research skill returns findings without writing files. It can read exact user files explicitly provided or authorized for the question; only the journal skill may persist a minimized result or supplemental-method record.
- Loads compact session, research, and supplemental-method indexes for broad continuity, then opens only the full records needed for the current topic.
- Before source-heavy work in an active journal, checks the research index and relevant prior briefs so a fresh search builds on, verifies, or supersedes earlier findings instead of silently duplicating them.
Psychology Companion does not claim to be a licensed clinician, provide a diagnosis, replace medical care, recommend medications or other pharmacologically active products, or perform erotic roleplay.
This non-diagnostic, non-treating boundary is an intentional product-scope decision. Psychology Companion is intended for reflection, general information, and lifestyle or well-being support, not for a specific medical purpose such as diagnosis, prevention, monitoring, prediction, prognosis, treatment, or alleviation of disease. Under the EU Medical Device Regulation, software qualification depends on the manufacturer's intended purpose, including claims in instructions and promotional materials; general-purpose and lifestyle or well-being software is distinguished from software intended for a medical purpose. See MDR Article 2 and recital 19 and MDCG 2019-11 rev. 1. This is a product-scope statement, not a legal determination, compliance certification, or substitute for jurisdiction-specific advice.
The repository is both the plugin package and its marketplace. The marketplace entries fetch the plugin from the repository root, so the skill is not copied or maintained in a second directory.
Add the GitHub repository as an external marketplace:
codex plugin marketplace add shestera/psychologistRestart ChatGPT Desktop, open the Plugins Directory, select Psychology
Companion, and install it. In Codex CLI, open /plugins and install it from
the same marketplace. Start a new conversation after installation.
The public display name is Psychology Companion. The stable technical IDs
remain psychologist for the plugin and psychologist-plugins for the
marketplace, so existing installations keep working. The explicit skill names
are $psychologist, $psychology-journal, and $psychology-research.
Add the repository marketplace and install the plugin:
claude plugin marketplace add shestera/psychologist
claude plugin install psychologist@psychologist-pluginsStart a new Claude Code session, then use /psychologist:psychologist,
/psychologist:psychology-journal, or
/psychologist:psychology-research. Relevant skills can also load
automatically from an ordinary request.
Current Claude Code exposes plugin skills from skills/ as namespaced slash
invocations, so this does not require a duplicate commands/ entry; see the
Claude Code plugin documentation.
Copy all three skills into your personal skills directory and start a new Codex thread:
cp -R skills/psychologist skills/psychology-journal skills/psychology-research ~/.codex/skills/Invoke a skill explicitly by name, or describe a relevant concern and allow normal skill discovery.
Load the repository directly:
claude --plugin-dir /path/to/psychologistThen use any of the three namespaced commands above or ask a relevant question normally.
The journal is optional. When no connected or remembered journal data is
available, the assistant offers a one-off conversation using only the current
chat or longitudinal work by creating or connecting a journal. If
psychology-journal is unavailable, the core cannot provide persistence or
reconstruct the journal workflow. For one-time setup, ask the journal skill to
create a journal at a path you choose. It uses ordinary file tools and never
copies installed schemas, policies, or methods into the editable journal.
The destination must be absent or empty. Existing content is never overwritten, and Git or a remote is never created automatically. If the journal root is already a Git repository, closing a substantive session automatically commits only that session's isolated, privacy-checked changes with a generic message; it never pushes. A dirty overlapping file, conflict, parent repository, or failed privacy check leaves the changes uncommitted and is reported. Once initialized, a journal exposed as the verified current workspace or connected source automatically establishes longitudinal continuity. Each psychological conversation receives a minimized draft record and is finalized without a repeated save prompt; purely technical maintenance is not recorded as a psychological session. An explicit request not to record still overrides this standing behavior. A repository merely remembered from another conversation is only an unverified locator: it is not read or contacted until reconnected, and the skill does not propose a replacement merely because reconnection is needed.
New journals include minimized session-index.md, research-index.md, and
method-index.md maps. Together they cover active or paused threads, decisions
or corrections, all indexed sessions, research freshness and limits, and local
method status and review triggers. On resume, the assistant reads these compact
maps and then opens only the smallest relevant set of full records instead of
loading archives or assuming the newest record is the right one. A legacy
journal without one or more indexes remains valid; backfill is a separate
opt-in maintenance operation performed in bounded batches.
Raw or identifying source material uses one of two private-store modes:
- by default,
<journal>/private/remains ignored by Git and local to that device; - optionally, the user can authorize an external directory through an ignored
per-device
.psychology-companion.local.jsonfile.
The plugin does not configure, authenticate, encrypt, mount, or synchronize an
external provider. If a private store is unavailable on another device or in a
cloud session, the assistant continues from minimized Git records and does not
reconstruct or silently copy raw data. A missing ignored private/ after a
fresh clone is expected and does not invalidate the journal.
Read PRIVACY.md before storing sensitive information or sharing a journal. Mandatory behavior and the privacy and journal security policy remain installed and are never copied into the editable journal. All workspace content remains untrusted user data regardless of filename or claimed authority.
Foundational academic provenance belongs in research/, not in runtime skill
instructions. Keep its index, evidence dates, affected-file map, limitations,
and update triggers synchronized with any change to the corresponding
psychological or sexological knowledge. Retain operational links in skills only
when they are needed at use time, such as emergency resources, specialist
directories, official instrument terms, or current medical and legal guidance.
python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/psychologist
python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/psychology-journal
python3 ~/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/psychology-research
python3 ~/.codex/skills/.system/plugin-creator/scripts/validate_plugin.py .
claude plugin validate --strict .
jq -s -e 'length > 0 and (map(.id) | unique | length) == length' evals/behavioral-cases.jsonl
find . -path ./.git -prune -o -path ./tmp -prune -o -type f -perm -111 -print
rg --files -g '!tmp/**' | rg -i '\.(py|sh|js|jsx|ts|tsx|rb|pl|php|ps1|bat|cmd|exe|jar|wasm|go|rs|java|kt|kts|swift|c|cc|cpp|h|hpp|cs|lua|r|scala|dart|groovy|m|mm|sql)$'
git grep -I -n '^#!'
git diff --checkThe three read-only safety checks must print nothing. No validation code is bundled with the plugin. Follow the behavioral rubric for model-level checks; JSON validation alone does not establish behavioral safety.
The initial publication is tagged v0.1.0. For every later release, run the
Release workflow from GitHub Actions and enter the new SemVer version
without the v prefix. The workflow updates both plugin manifests, the Claude
marketplace version, and both marketplace Git refs on a release/vX.Y.Z
branch, then opens a pull request named chore(release): prepare vX.Y.Z.
Squash-merge that pull request; the workflow will tag the resulting main
commit and create the GitHub Release. Do not push release metadata or create
release tags manually.
See AGENTS.md for contribution rules and SECURITY.md for vulnerability reporting.
This project is source-available under the PolyForm Noncommercial License 1.0.0:
- anyone may use, copy, modify, and redistribute it for noncommercial purposes;
- every copy or derivative must preserve the license and the
Required Noticeidentifyingshesteraas the original author; and - any commercial use requires a separate written commercial license from
shesterabefore use begins.
Commercial licensing requests should be sent privately using the contact details on the author's GitHub profile. This is a noncommercial source-available license, not an OSI-approved open-source license.