get-shit-done/workflows/profile-user.md
This workflow wires Phase 1 (session pipeline) and Phase 2 (profiling engine) into a cohesive user-facing experience. All heavy lifting is done by existing gsd-sdk query handlers (with legacy gsd-tools.cjs parity where needed) and the gsd-user-profiler agent -- this workflow orchestrates the sequence, handles branching, and provides the UX.
</purpose>
<required_reading> Read all files referenced by the invoking prompt's execution_context before starting.
Key references:
Parse flags from $ARGUMENTS:
--questionnaire flag (skip session analysis, questionnaire-only)--refresh flag (rebuild profile even when one exists)Check for existing profile:
PROFILE_PATH="$HOME/.claude/get-shit-done/USER-PROFILE.md"
[ -f "$PROFILE_PATH" ] && echo "EXISTS" || echo "NOT_FOUND"
If profile exists AND --refresh NOT set AND --questionnaire NOT set:
Text mode (workflow.text_mode: true in config or --text flag): Set TEXT_MODE=true if --text is present in $ARGUMENTS OR text_mode from init JSON is true. When TEXT_MODE is active, replace every AskUserQuestion call with a plain-text numbered list and ask the user to type their choice number. This is required for non-Claude runtimes (OpenAI Codex, Gemini CLI, etc.) where AskUserQuestion is not available.
Use AskUserQuestion:
If "View it": Read USER-PROFILE.md, display its content formatted as a summary card, then exit. If "Refresh it": Set --refresh behavior and continue. If "Cancel": Display "No changes made." and exit.
If profile exists AND --refresh IS set:
Backup existing profile:
cp "$HOME/.claude/get-shit-done/USER-PROFILE.md" "$HOME/.claude/USER-PROFILE.backup.md"
Display: "Re-analyzing your sessions to update your profile." Continue to step 2.
If no profile exists: Continue to step 2.
Skip if --questionnaire flag is set (no JSONL reading occurs -- jump directly to step 4b).
Display consent screen:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GSD > PROFILE YOUR CODING STYLE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Claude starts every conversation generic. A profile teaches Claude
how YOU actually work -- not how you think you work.
## What We'll Analyze
Your recent Claude Code sessions, looking for patterns in these
8 behavioral dimensions:
| Dimension | What It Measures |
|----------------------|---------------------------------------------|
| Communication Style | How you phrase requests (terse vs. detailed) |
| Decision Speed | How you choose between options |
| Explanation Depth | How much explanation you want with code |
| Debugging Approach | How you tackle errors and bugs |
| UX Philosophy | How much you care about design vs. function |
| Vendor Philosophy | How you evaluate libraries and tools |
| Frustration Triggers | What makes you correct Claude |
| Learning Style | How you prefer to learn new things |
## Data Handling
✓ Reads session files locally (read-only, nothing modified)
✓ Analyzes message patterns (not content meaning)
✓ Stores profile at $HOME/.claude/get-shit-done/USER-PROFILE.md
✗ Nothing is sent to external services
✗ Sensitive content (API keys, passwords) is automatically excluded
If --refresh path: Show abbreviated consent instead:
Re-analyzing your sessions to update your profile.
Your existing profile has been backed up to USER-PROFILE.backup.md.
Use AskUserQuestion:
If default (no --refresh) path:
Use AskUserQuestion:
Display: "◆ Scanning sessions..."
Run session scan:
SCAN_RESULT=$(gsd-sdk query scan-sessions --json 2>/dev/null)
Parse the JSON output to get session count and project count.
Display: "✓ Found N sessions across M projects"
Determine data sufficiency:
Display: "◆ Sampling messages..."
Run profile sampling:
SAMPLE_RESULT=$(gsd-sdk query profile-sample --json 2>/dev/null)
Parse the JSON output to get the temp directory path and message count.
Display: "✓ Sampled N messages from M projects"
Display: "◆ Analyzing patterns..."
Spawn gsd-user-profiler agent using Task tool:
Use the Task tool to spawn the gsd-user-profiler agent. Provide it with:
$HOME/.claude/get-shit-done/references/user-profiling.mdThe agent prompt should follow this structure:
Read the profiling reference document and the sampled session messages, then analyze the developer's behavioral patterns across all 8 dimensions.
Reference: @$HOME/.claude/get-shit-done/references/user-profiling.md
Session data: @{temp_dir}/profile-sample.jsonl
Analyze these messages and return your analysis in the <analysis> JSON format specified in the reference document.
Parse the agent's output:
<analysis> JSON block from the agent's responseANALYSIS_PATH="{temp_dir}/analysis.json"
Write the analysis JSON to $ANALYSIS_PATH.
Display: "✓ Analysis complete (N dimensions scored)"
Check for thin data:
Continue to step 5.
Display: "Using questionnaire to build your profile."
Get questions:
QUESTIONS=$(gsd-sdk query profile-questionnaire --json 2>/dev/null)
Parse the questions JSON. It contains 8 questions, one per dimension.
Present each question to the user via AskUserQuestion:
For each question in the questions array:
Collect all answers into an answers JSON object mapping dimension keys to selected answer values.
Save answers to temp file:
ANSWERS_PATH=$(mktemp /tmp/gsd-profile-answers-XXXXXX.json)
Write the answers JSON to $ANSWERS_PATH.
Convert answers to analysis:
ANALYSIS_RESULT=$(gsd-sdk query profile-questionnaire --answers "$ANSWERS_PATH" --json 2>/dev/null)
Parse the analysis JSON from the result.
Save analysis JSON to a temp file:
ANALYSIS_PATH=$(mktemp /tmp/gsd-profile-analysis-XXXXXX.json)
Write the analysis JSON to $ANALYSIS_PATH.
Continue to step 5 (skip split resolution since questionnaire handles ambiguity internally).
Skip if questionnaire-only path (splits already handled internally).
Read the analysis JSON from $ANALYSIS_PATH.
Check each dimension for cross_project_consistent: false.
For each split detected:
Use AskUserQuestion:
If user picks a specific rating: Update the dimension's rating field in the analysis JSON to the selected value.
If user picks "Context-dependent": Keep the dominant rating in the rating field. Add a context_note to the dimension's summary describing the split (e.g., "Context-dependent: terse in CLI projects, detailed in frontend projects").
Write updated analysis JSON back to $ANALYSIS_PATH.
Display: "◆ Writing profile..."
gsd-sdk query write-profile --input "$ANALYSIS_PATH" --json
Display: "✓ Profile written to $HOME/.claude/get-shit-done/USER-PROFILE.md"
Read the analysis JSON from $ANALYSIS_PATH to build the display.
Show report card table:
## Your Profile
| Dimension | Rating | Confidence |
|----------------------|----------------------|------------|
| Communication Style | detailed-structured | HIGH |
| Decision Speed | deliberate-informed | MEDIUM |
| Explanation Depth | concise | HIGH |
| Debugging Approach | hypothesis-driven | MEDIUM |
| UX Philosophy | pragmatic | LOW |
| Vendor Philosophy | thorough-evaluator | HIGH |
| Frustration Triggers | scope-creep | MEDIUM |
| Learning Style | self-directed | HIGH |
(Populate with actual values from the analysis JSON.)
Show highlight reel:
Pick 3-4 dimensions with the highest confidence and most evidence signals. Format as:
## Highlights
- **Communication (HIGH):** You consistently provide structured context with
headers and problem statements before making requests
- **Vendor Choices (HIGH):** You research alternatives thoroughly -- comparing
docs, GitHub activity, and bundle sizes before committing
- **Frustrations (MEDIUM):** You correct Claude most often for doing things
you didn't ask for -- scope creep is your primary trigger
Build highlights from the evidence array and summary fields in the analysis JSON. Use the most compelling evidence quotes. Format each as "You tend to..." or "You consistently..." with evidence attribution.
Offer full profile view:
Use AskUserQuestion:
Use AskUserQuestion with multiSelect:
If no artifacts selected: Display "No artifacts generated. Your profile is saved at $HOME/.claude/get-shit-done/USER-PROFILE.md" and jump to step 10.
Generate selected artifacts sequentially (file I/O is fast, no benefit from parallel agents):
For /gsd-dev-preferences (if selected):
gsd-sdk query generate-dev-preferences --analysis "$ANALYSIS_PATH" --json
Display: "✓ Generated /gsd-dev-preferences at $HOME/.claude/skills/gsd-dev-preferences/SKILL.md"
For CLAUDE.md profile section (if selected):
gsd-sdk query generate-claude-profile --analysis "$ANALYSIS_PATH" --json
Display: "✓ Added profile section to CLAUDE.md"
For Global CLAUDE.md (if selected):
gsd-sdk query generate-claude-profile --analysis "$ANALYSIS_PATH" --global --json
Display: "✓ Added profile section to $HOME/.claude/CLAUDE.md"
Error handling: If any gsd-sdk query or gsd-tools.cjs call fails, display the error message and use AskUserQuestion to offer "Retry" or "Skip this artifact". On retry, re-run the command. On skip, continue to next artifact.
If --refresh path:
Read both old backup and new analysis to compare dimension ratings/confidence.
Read the backed-up profile:
BACKUP_PATH="$HOME/.claude/USER-PROFILE.backup.md"
Compare each dimension's rating and confidence between old and new. Display diff table showing only changed dimensions:
## Changes
| Dimension | Before | After |
|-----------------|-----------------------------|-----------------------------|
| Communication | terse-direct (LOW) | detailed-structured (HIGH) |
| Debugging | fix-first (MEDIUM) | hypothesis-driven (MEDIUM) |
If nothing changed: Display "No changes detected -- your profile is already up to date."
Display final summary:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
GSD > PROFILE COMPLETE ✓
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Your profile: $HOME/.claude/get-shit-done/USER-PROFILE.md
Then list paths for each generated artifact:
Artifacts:
✓ /gsd-dev-preferences $HOME/.claude/skills/gsd-dev-preferences/SKILL.md
✓ CLAUDE.md section ./CLAUDE.md
✓ Global CLAUDE.md $HOME/.claude/CLAUDE.md
(Only show artifacts that were actually generated.)
Clean up temp files:
Remove the temp directory created by profile-sample (contains sample JSONL and analysis JSON):
rm -rf "$TEMP_DIR"
Also remove any standalone temp files created for questionnaire answers:
rm -f "$ANSWERS_PATH" 2>/dev/null
rm -f "$ANALYSIS_PATH" 2>/dev/null
(Only clean up temp paths that were actually created during this workflow run.)
</process><success_criteria>