gse-hug

name	gse-hug
description	Establish or update user profile. Triggered when user context is unknown or /gse:hug is called.

GSE-One HUG — User Profile

Arguments: $ARGUMENTS

Options

Flag / Sub-command	Description
(no args)	Start or resume the user profile interview
--update	Update an existing profile (re-ask only changed dimensions)
--show	Display the current profile without modification
--team	Enable team mode (per-user profiles)
--help	Show this command's usage summary

Prerequisites

Before executing, read:

1. .gse/profile.yaml — existing profile (if any)
2. .gse/config.yaml — project configuration (if any)
3. pyproject.toml / package.json / Cargo.toml — project manifest for domain inference
4. Git log — for team context inference

Workflow

Step 0 — Language Detection (always first, always alone)

` end marker is present. -->

Condition: No language is set in .gse/profile.yaml or .gse/config.yaml.

If a language is already configured, skip to Step 1.

1. Detect system locale — Read $LANG or $LC_ALL to identify the system language (e.g., fr_FR.UTF-8 → French).

2. Build the option list — Start with the detected locale language (marked "Recommended"), then add the standard list: English, Français, Español, Deutsch. Deduplicate: if the detected locale matches one of the four, move it to first position instead of duplicating.

3. Ask in all listed languages — The question itself is displayed in each language so the user can understand it regardless of their language:

   Example (interactive — detected locale = Japanese). Use your runtime's native interactive question tool (see orchestrator P4 — AskUser maps to AskUserQuestion on Claude Code, AskQuestion on Cursor ≥2.4, question on opencode). Apply the template literally:

   AskUser([
     { question: "Which language would you like me to use? / どの言語を使いますか？ / Quelle langue souhaitez-vous ? / ¿Qué idioma prefiere? / Welche Sprache bevorzugen Sie?",
       header: "Language",
       multiSelect: false,
       options: [
         { label: "日本語 (Recommended)", description: "Detected from your system / システムから検出" },
         { label: "English", description: "English" },
         { label: "Français", description: "French" },
         { label: "Español", description: "Spanish" }
       ] }
   ])
   // "Other" is automatically available — user can type Deutsch, 中文, हिन्दी, etc.
   

   Example (text fallback — detected locale = French):

   🌍 Which language would you like me to use?
      Quelle langue souhaitez-vous utiliser ?
      ¿Qué idioma desea utilizar?
      Welche Sprache möchten Sie verwenden?
   
   1. Français (Recommended — detected from your system)
   2. English
   3. Español
   4. Deutsch
   5. Other — type your language
   

4. Set artifact language — After the user chooses the chat language, set language.artifacts to en (industry default). Inform the user: "I'll chat in [chosen language]. Files I produce (code, docs) will be in English by default — you can change this anytime." This covers the artifacts and overrides sub-fields of dimension #4, which are NOT re-asked during Step 2.

5. Switch immediately — From this point on, ALL communication is in the chosen language.

Step 1 — Smart Inference

Before asking any questions, infer as much as possible from available signals:

Dimension	Inference Source
Project domain	Package manifest, README, directory structure
IT expertise	Vocabulary complexity in user messages (e.g., "deploy" vs "put online")
Team context	Git log — number of contributors, commit patterns
Scientific expertise	Presence of academic references, LaTeX files, data science libraries

Report inferences (in the user's chosen language): "Based on what I can see, I'll assume: [inferred values]. I'll only ask about what I can't determine."

Step 2 — Interview (Only Unresolved Dimensions)

The 13 HUG dimensions are:

#	Dimension	Scale / Values	Purpose
1	IT expertise	beginner / intermediate / advanced / expert	Calibrate technical depth of explanations
2	Scientific expertise	none / familiar / practitioner / researcher	Adjust formality and rigor expectations
3	Abstraction capability	concrete-first / balanced / abstract-first	Choose explanation style (examples vs theory)
4	Language	chat: ISO 639-1, artifacts: ISO 639-1, overrides: per-type	Set chat language, artifact language, and per-type overrides. Default: chat=detected, artifacts=en. Tell the user: "I'll communicate in [chat language]. Files will be produced in [artifacts language] by default. You can change this at any time, globally or per document."
5	Preferred verbosity	terse / normal / detailed	Control output length and ceremony level
6	Domain background	free text	Tailor domain-specific vocabulary and examples
7	Decision involvement	autonomous / collaborative / supervised	Control Gate frequency and agent autonomy
8	Project domain	web / api / cli / data / mobile / embedded / library / scientific / other	Calibrate default tech stack and test pyramid
9	Team context	solo / pair / small-team / large-team	Adjust collaboration ceremonies
10	Learning goals	list of short goal strings (optional)	Activate proactive LEARN at relevant moments. Three entry points into learning even if empty: (a) explicit user invocation /gse:learn <topic>, (b) coach proactive gap detection (default: on — see plugin/agents/coach.md pedagogy axis), (c) retrospective proposal at /gse:compound Axe 3 based on sprint signals. Leaving the list empty is a valid choice; it does not disable learning.
11	Contextual tips	on / off	Enable/disable inline micro-explanations
12	Emoji	on / off	Enable/disable emoji in chat output (default: on)
13	User name	free text (optional)	Display name in dashboard and artefacts. For beginners: "What name should I use for you in the project? (You can skip this.)" Store in profile.yaml under user.name. If skipped, defaults to git user name or "Unknown".

Question cadence (adaptive)

The number of questions asked at once is proportional to the user's IT expertise. IT expertise is the cadence key: the first question after Step 0 is the IT expertise level, asked alone, UNLESS it was confidently inferred at Step 1 and confirmed via the inference report, in which case the inferred value sets the cadence for all remaining questions.

IT Expertise	Cadence	Rationale
Beginner	1 question at a time — wait for response before asking the next	Avoid cognitive overload; let the user absorb each concept
Intermediate	2-3 questions grouped by related theme	Comfortable enough to handle a small batch
Advanced / Expert	All remaining questions in one block (interactive widget only — see "Text-fallback discipline" below for the no-widget case)	Efficient; no hand-holding needed

Interview sequence:

1. IT expertise — the cadence key: asked alone first after Step 0, UNLESS confidently inferred at Step 1 and confirmed via the inference report, in which case the inferred value sets the cadence
2. Remaining dimensions — asked at the cadence determined by Step 2.1

Completeness invariant. Every dimension NOT resolved by Step 1 — Smart Inference MUST be surfaced to the user as an explicit numbered choice before the profile is saved. A "recommended preset" or bundled default is NOT a substitute for asking — a silently-defaulted preference dimension is a defect, not a shortcut (spec §14.1 — Sprint 1 vs Sprint N+1: "HUG = Full interview (all 13 dimensions)"). This holds especially for the consent-bearing dimensions — #7 Decision involvement, #10 Learning goals, #11 Contextual tips — which shape how autonomously the agent acts and whether it teaches as it works (§P14 — Knowledge Transfer); these are never inferred and never preset-defaulted. (Rule added after a Codex onboarding without the interactive widget improvised a preset that set contextual tips ON without ever asking — the user never got to choose the pedagogical behavior.)

Text-fallback discipline (no interactive widget — e.g. Codex / Gemini CLI, see spec Appendix C — Per-platform session-transcript persistence's sibling era-bound rows on the absent interactive-question widget). When the agent falls back to text (spec §P4 — Human-in-the-Loop, "Text fallback"), the "all in one block" expert cadence MUST NOT become a single composite message with free-text fields — that maximizes keystrokes, the opposite of the intent. In text fallback, at every expertise level:

• present each finite-option dimension as a numbered list, answerable by a single number;
• group at most ~3 dimensions per message (even for experts), so the user replies with a short list like 2 4 1, never a paragraph;
• convert otherwise-free-text dimensions to numbered inferred guesses + an "other: type…" escape — e.g. Domain background offers the top inferred domains as numbered options, free text only as the explicit escape.

This mirrors go.md Step 7 — Intent Capture, which is already keystroke-minimal.

Interactive mode (preferred): When the environment provides an interactive question tool (all three supported runtimes do — see orchestrator P4 for the AskUser mapping table), use it to present profile questions as clickable choices instead of numbered text lists. Use multiSelect: true for dimensions where multiple values apply. Map "Discuss" to the automatic "Other" escape hatch. When interactive tools are unavailable or dimensions require free text (e.g., learning goals, domain background), fall back to the numbered-list format (spec §P4 — Human-in-the-Loop, "Text fallback"), following the "Text-fallback discipline" below — not a free-form conversational block.

Example — beginner flow (1 question at a time):

[Step 0] → Language selected: English
[Step 1] → Inferences reported in English
[Step 2.1] → "How would you describe your experience with computer tools?"
             → Response: beginner → cadence = 1 at a time
[Step 2.2] → "Do you prefer short answers or detailed explanations?"
             → Response
[Step 2.3] → "Would you like me to make decisions on my own, or consult you first?"
             → Response
[Step 2.4] → ... (1 question at a time until complete)

Example — expert flow (all at once):

[Step 0] → Language question asked (always — locale-detected default preselected)
[Step 1] → Inferences: IT=expert (from vocabulary), domain=api, team=solo
[Step 2]  → All remaining questions in one block:
            AskUser([
              { question: "Verbosity?", ... },
              { question: "Decision involvement?", ... },
              { question: "Learning goals?", ... }
            ])

Step 3 — Team Mode

When --team is specified or multiple git contributors are detected:

1. Detect git user — Read git config user.name and git config user.email
2. Check profiles directory — Look for .gse/profiles/{username}.yaml
3. Load or create — If profile exists, load it. If not, run the interview for this user.
4. Link — Copy the active user's profile to .gse/profile.yaml (or create a symlink on systems that support it)
5. Switch — When git user changes between sessions, auto-switch the active profile by updating .gse/profile.yaml

Profile file structure:

.gse/
├── profile.yaml          # active profile (copy of profiles/nicolas.yaml)
└── profiles/
    ├── nicolas.yaml
    └── alex.yaml

Step 4 — Git Initialization

Verify the project environment is ready. This step is blocking — do NOT mention /gse:go or any next steps until git initialization is fully resolved (accepted or declined).

1. Git repo check — If no .git/ directory exists:

   • For beginner users (per P9 — Adaptive Communication): Explain what git is before asking, and warn about the system dialog that will appear:

     Agent: Before we continue, I need to set up version control for your project.
     
     Version control (git) is like an "infinite undo" — it keeps a complete 
     history of every change you make, so you can always go back to a previous 
     version if something goes wrong.
     
     If you agree, I'll initialize it now. A confirmation dialog will appear 
     in the editor — click "Run" to confirm (or "Skip" to cancel).
     
     Should I set it up?
     

   • For intermediate/advanced users: Ask concisely: "No git repo detected here. Should I run git init?"

   • Wait for response — Do NOT proceed or mention next steps until the user responds.

   • If yes: run git init, create initial .gitignore, apply the Git Identity Verification preflight (see below), then create a foundational commit on main:

     git init
     # create .gitignore: seed it from the GSE-One additions template, then add project-appropriate entries
     cat "$([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/templates/gitignore-additions.txt" >> .gitignore
     # [Git Identity Verification preflight — see below]
     git add .gitignore
     git commit -m "chore: initialize repository"
     

     This foundational commit is mandatory — without it, main is not a valid branch reference and all subsequent branching operations (P12) will fail. If a system permission dialog appears, the user has already been told what to do.

     Git Identity Verification preflight (Hard guardrail, spec P12.6): Before running git commit, verify that a git identity is configured.

     1. Check git installation — Run git --version. If exit code is non-zero (git not installed), abort with: "git is not installed on this system. GSE-One requires git — please install it first." No Gate is shown.

     2. Query identity at both scopes:

        • git config --global user.name and git config --global user.email
        • git config --local user.name and git config --local user.email

     3. If both name AND email are set at either scope (global OR local) → identity is OK, proceed to the commit.

     4. If any value is missing → present the Git Identity Gate:

        Question: I need a git identity to save commits — none is set on this machine, globally or for this project. How should I proceed?

        Options:

        1. Set global identity (recommended) — I'll configure git with your name and email, so all your projects on this machine can commit. I'll ask for your name and email.
        2. Set local identity (this project only) — I'll configure git for this project only. Your global setup is unchanged. I'll ask for your name and email.
        3. Quick placeholder (local, for a quick test) — I'll set GSE User / user@local for this project only. Good for throwaway experiments; change later with options 1 or 2.
        4. I'll set it myself — Here are the commands to run in your terminal:

           git config --global user.name "Your Name"
           git config --global user.email "[e-mail]"
           

           Let me know when you're done — I'll re-check and continue.
        5. Discuss — Explain the scope difference and security considerations.

        Beginner translations (per P9 Adaptive Communication): option 1 → "set up your signature for all your projects"; option 2 → "just for this folder"; option 3 → "a disposable signature for this quick test"; option 4 → "I'll give you two commands to type yourself".

     5. On option 1 or 2: ask for the user's name, then ask for the email (one at a time for beginners, grouped for intermediate/expert). Validate the email (must contain exactly one @ with a non-empty local part and a dotted domain with non-empty labels on both sides of the dot — e.g. [e-mail]). On invalid email, re-prompt: "That doesn't look like a valid email address. Could you enter it again? (e.g., [e-mail])". Once valid, run:

        • Option 1: git config --global user.name "<name>" and git config --global user.email "<email>"
        • Option 2: git config --local user.name "<name>" and git config --local user.email "<email>" Re-verify identity, then proceed to the commit.

     6. On option 3: run git config --local user.name "GSE User" and git config --local user.email "user@local". Immediately print a single reminder: "Placeholder identity set locally for this project. If you plan to share this repo or push to a remote, replace it via /gse:hug --update or directly with git config --global user.name / user.email." Do not repeat the reminder on subsequent activities. Proceed to the commit.

     7. On option 4: print the two git config --global ... commands as a copy-paste block, wait for user confirmation ("done" / "ok"), then re-run the identity detection. If still missing, re-present the Gate (options 1-5).

     8. On option 5: explain (scope difference between global/local, visibility of commits on pushed branches, throwaway nature of option 3), then re-present options 1-4.

   • If no: acknowledge and continue without git. Note that some GSE features (branching, version control guardrails) will be unavailable.

2. Create .gse/ directory — If it does not exist:
   • Create .gse/ with subdirectories: profiles/, checkpoints/
   • Copy config.yaml from the template ($([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/templates/config.yaml) with default values. Leave lifecycle.mode empty — /gse:go will fill it after Complexity Assessment (spec §14.3 — Orchestrator Decision Logic, Step 6 — Complexity Assessment).
   • Copy status.yaml from the template ($([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/templates/status.yaml) verbatim. Its canonical initial state is current_sprint: 0 / current_phase: LC01 — the pre-first-sprint baseline (PLAN promotes to sprint 1; see go.md Adopt Step 5 — Set baseline). Do NOT hand-author the file with null cursor values: a null current_sprint is the legitimate baseline meaning but must be encoded as 0, not null, so the dashboard and downstream activities read a numeric sprint.
   • Stamp the GSE-One plugin version in status.yaml.gse_version by reading it from the unified version file at $([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/VERSION — a single 1-line text file present at the root of every install target (regardless of platform or plugin/no-plugin mode), written by the generator and distributed by install.py. This traces which plugin version created the project state — used by the dashboard header display and by future /gse:upgrade migration logic (spec §13.4 — Required fields, design §14.3 — Open questions #3). See CLAUDE.md "Tool architecture" for the registry model that makes this path uniformly resolvable.
   • Add to .gitignore: entries for .gse/local/ (machine-specific state)
3. Save profile — Write the completed profile to .gse/profile.yaml (or .gse/profiles/{username}.yaml in team mode)

Step 4.5 — Update Mode (--update)

When called with --update, the profile already exists. Only re-ask dimensions the user wants to change:

1. Load current profile — Read .gse/profile.yaml and display current values as a summary.

2. Ask what to change — Present all dimensions with their current values. The user selects which ones to modify.

3. Re-interview selected dimensions — Run Step 2 interview only for the selected dimensions.

4. Detect behavioral impact — Compare old and new values. If any of the following dimensions changed, notify the user of the consequences:

   Changed dimension	Impact	Notification
   it_expertise	Vocabulary, question cadence, artefact approval, knowledge transfer	"I'll adjust how I explain things and how many questions I ask at once."
   language.chat	All chat output language	"From now on, I'll communicate in [new language]."
   decision_involvement	Gate frequency, supervised mode override	"I'll [ask more / ask less] before making technical choices."
   preferred_verbosity	Output length and detail level	"I'll make my responses [shorter / more detailed]."
   contextual_tips	Inline micro-explanations	"I'll [start / stop] adding learning tips during activities."

5. Save and apply immediately — Update profile.yaml with the new values and set updated timestamp. The changes take effect on the very next skill invocation — no restart needed.

Step 5 — Profile Persistence

Save the profile as YAML:

# .gse/profile.yaml
version: 1
created: 2026-01-15
updated: 2026-01-15
user:
  name: "Nicolas"
  git_email: "[e-mail]"
dimensions:
  it_expertise: advanced
  scientific_expertise: practitioner
  abstraction_capability: balanced
  language:
    chat: fr
    artifacts: en
    overrides: {}
  preferred_verbosity: normal
  domain_background: "Software engineering, AI-assisted development"
  decision_involvement: collaborative
  project_domain: api
  team_context: solo
  learning_goals:
    - "Rust async patterns"
    - "property-based testing"
  contextual_tips: true
  emoji: true
inferred:
  it_expertise: true
  project_domain: true
  team_context: true
  language.chat: true

Step 5.5 — Dashboard Initialization

After .gse/ is created and profile saved:

1. Validate tool access — Run: [ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one to verify the plugin registry exists. If the file is missing, warn the user: "GSE-One registry not found. Run python3 install.py again to fix."
2. Generate the first dashboard — Run: python3 "$([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/tools/dashboard.py"
3. Verify — Check that docs/dashboard.html was created and is not empty.
4. Inform the user:
   • For beginner users: "I've created a dashboard to track your project's progress. You can open it in your browser at any time: docs/dashboard.html. It updates at each step." (adapt wording to the user's language)
   • For intermediate/advanced users: "Dashboard generated at docs/dashboard.html. Regenerated after each activity. Run python3 \"$([ -s .gse/registry ] && cat .gse/registry || cat ~/.gse-one)/tools/dashboard.py\" --watch for live updates."

Step 6 — Transition

Only after Steps 4-5.5 are fully completed (git initialization resolved, profile saved, dashboard created):

• If project has no .gse/config.yaml: propose /gse:go to start project setup.
  • For beginner users: Explain what /gse:go does before suggesting it: "Now that your profile is saved, the next step is to set up your project. Type /gse:go — it will help you define what you want to build and create a first work plan."
  • For intermediate/advanced users: "Profile saved. Run /gse:go to set up project state (config.yaml, first sprint)."
• If project already configured: confirm profile saved and return to previous activity