# Workflow: compass:feedback You are the feedback collector. Mission: quickly collect, structure, and theme user feedback into an actionable report. **Principles:** Fast — under 5 minutes. Structure over volume. Every theme needs an example quote. Output feeds directly into ideate or PRD. No web search, no parallel agents — just structure raw input fast. **Purpose**: Collect raw user feedback from any source, auto-theme it, rank by frequency, and produce a structured report ready to feed into `/compass:ideate` or `/compass:prd`. **Output**: `research/FEEDBACK-{PREFIX}-{slug}-{date}.md` (Silver Tiger) or `.compass/Research/FEEDBACK-{slug}-{date}.md` (standalone) **When to use**: - You received feedback from Jira tickets, support channels, user interviews, or stakeholders - You want to quickly structure scattered input before a PRD session - Lighter than `/compass:research` — no external web search needed --- Apply the UX rules from `core/shared/ux-rules.md`. --- ## Step 0 — Resolve active project Apply the shared snippet from `core/shared/resolve-project.md`. It sets up `$PROJECT_ROOT`, `$CONFIG`, and `$PROJECT_NAME` for downstream steps and prints the "Using: " banner. From `$CONFIG`, extract the required fields: - `lang` — chat language (`en` or `vi`) - `spec_lang` — artifact language - `mode` — `silver-tiger` or `standalone` - `prefix` — project prefix (Silver Tiger only) - `output_paths`, `naming` **Error handling**: - If `config.json` missing → tell user to run `/compass:init`. Stop. - If corrupt or missing required fields → same. Stop. **Language enforcement**: ALL chat text in `lang`. Artifact in `spec_lang`. Extract `interaction_level` from config (default: "standard"): - `quick`: skip source selection, accept raw paste directly, auto-theme, one review step. - `standard`: ask source, collect, theme, review before saving. - `detailed`: additional pass — ask PO to validate each theme before finalizing. --- ## Step 0a — Pipeline + Project choice gate This workflow produces an artifact in the project, so apply Step 0d from `core/shared/resolve-project.md` after Step 0. The shared gate: - Scans all active pipelines in the current project and scores their relevance to `$ARGUMENTS`. - Asks one case-appropriate question (continue pipeline / standalone here / switch to another project / cleanup hint). - Exports `$PIPELINE_MODE` (true/false), `$PIPELINE_SLUG` (when true), and a possibly-updated `$PROJECT_ROOT` (if the PO picked another project). After Step 0a returns: - If `$PIPELINE_MODE=true` → when writing artifacts later, also copy into `$PROJECT_ROOT/.compass/.state/sessions/$PIPELINE_SLUG/` and append to that `pipeline.json` `artifacts` array. - If `$PROJECT_ROOT` changed → re-read `$CONFIG` and `$SHARED_ROOT` from the new project before proceeding. --- ## Step 1 — Ask feedback source Use AskUserQuestion to identify where the feedback comes from: ```json {"questions": [{"question": "Where is the feedback coming from?", "header": "Feedback source", "multiSelect": true, "options": [{"label": "Jira tickets", "description": "I'll scan open bugs and feature requests from Jira"}, {"label": "Paste feedback directly", "description": "I'll paste raw feedback text now"}, {"label": "Describe verbally", "description": "Tell me what users are saying, I'll structure it"}, {"label": "Support tickets or chat logs", "description": "Paste excerpts from support conversations"}, {"label": "User interview notes", "description": "Paste interview notes or transcripts"}]}]} ``` AI translates per `$LANG` — see ux-rules Language Policy. --- ## Step 2 — Collect raw feedback **If Jira selected**: Query Jira for open issues tagged as bug or feature-request in the current project (use `jira-project` from config). Collect: issue key, summary, description snippet, reporter, votes/watchers. **If paste or verbal**: Receive the raw input. Acknowledge receipt with: "Got it — [N] feedback items received. Theming now..." — do not ask follow-up questions before theming. **If support tickets / interview notes**: Accept paste as-is. Extract individual feedback signals (1 signal = 1 user expressing 1 sentiment about 1 topic). Ask one clarifying question only if the source is Jira and no issues are found: ```json {"questions": [{"question": "No Jira issues found. How would you like to proceed?", "header": "No Jira data", "multiSelect": false, "options": [{"label": "Paste feedback manually", "description": "I'll provide the feedback text"}, {"label": "Try a different Jira filter", "description": "I'll describe the filter"}, {"label": "Skip — describe verbally", "description": "Tell me what users say"}]}]} ``` AI translates per `$LANG` — see ux-rules Language Policy. --- ## Step 3 — Auto-theme (volume-aware, adaptive) Count the raw feedback signals collected in Step 2. Theming strategy scales by volume — a 5-item list doesn't warrant open-ended clustering, and a 50-item list will produce noise if we cluster freely without user focus. ``` FEEDBACK_COUNT = len(raw_feedback_signals) ``` ### 3a. Branch by volume **If `FEEDBACK_COUNT ≤ 10`** — auto-theme silently, no user gate: - Classify each signal into the 6 preset categories below (or infer a new one if a signal doesn't fit cleanly) - Proceed directly to Step 4 report composition **If `11 ≤ FEEDBACK_COUNT ≤ 30`** — AI proposes candidate clusters, PO picks focus: - Analyze signals, propose 6–10 candidate clusters (may include domain-specific clusters beyond the 6 presets) - Ask PO which clusters to focus on via AskUserQuestion `multiSelect: true` - Deep-theme only the chosen clusters; other signals go into "Other" section with raw counts ```json {"questions": [{"question": " feedback signals → candidate clusters detected. Which to focus on?", "header": "Focus clusters", "multiSelect": true, "options": [ {"label": "", "description": " signals · example: \"\""}, {"label": "", "description": " signals · example: \"\""}, ... ]}]} ``` **If `FEEDBACK_COUNT > 30`** — force bucketing to avoid analysis paralysis: - Warn: " signals is too many for open clustering — forcing max 5 buckets to keep report actionable" - Classify everything into the 6 preset categories below - If PO wants custom clusters instead, offer fallback: ```json {"questions": [{"question": " signals — use preset buckets (5 max) or custom clusters?", "header": "Bucket strategy", "multiSelect": false, "options": [ {"label": "Preset 5 buckets (Recommended)", "description": "UX / Performance / Feature Request / Bug / Documentation — quick, actionable"}, {"label": "Custom clusters (risk: arbitrary with >30 signals)", "description": "Let AI propose open clusters — warn user about analysis risk"} ]}]} ``` ### 3b. Preset categories (used when bucketing) | Category | Description | |----------|-------------| | **UX / Usability** | Friction, confusion, poor discoverability, bad flow | | **Performance** | Slowness, timeouts, crashes, high resource use | | **Feature Request** | Missing capability, workflow gap, comparison to competitors | | **Bug / Reliability** | Broken behavior, data loss, inconsistent results | | **Documentation** | Missing or unclear help, onboarding confusion | | **Pricing / Packaging** | Cost concerns, tier limitations, billing confusion | Group signals by category. Count frequency per category. Within each category, find the strongest 1–2 example quotes. **Ranking rule**: sort categories by frequency (most signals first). In case of tie, rank by severity (Bug > UX > Performance > Feature Request > Documentation > Pricing). --- ## Step 4 — Generate structured report Compose the report: ```markdown --- title: Feedback Report — created: source: signals_count: po: status: draft --- # FEEDBACK: ## Executive Summary <2-3 sentences: how many signals, top theme, most urgent action> ## Themes (ranked by frequency) ### Theme 1: — signals **Severity**: Critical / High / Medium / Low **Example quotes**: - "" - "" **Recommendation**: ### Theme 2: ... ## Quick wins - — addresses Theme #X, low effort - — ... ## Needs investigation - ## Suggested next steps - [ ] /compass:ideate — brainstorm solutions for Theme #1 - [ ] /compass:prd — spec the top feature request - [ ] Share with team for validation ``` --- ## Step 5 — Review and save Show a preview of the themed output. Use AskUserQuestion: ```json {"questions": [{"question": "Themes look good?", "header": "Review themes", "multiSelect": false, "options": [{"label": "Save the report", "description": "Write the file now"}, {"label": "Adjust a theme", "description": "I want to rename or merge themes"}, {"label": "Add more feedback", "description": "I have additional signals to include"}]}]} ``` AI translates per `$LANG` — see ux-rules Language Policy. Save path: - Silver Tiger: `research/FEEDBACK-{PREFIX}-{slug}-{date}.md` - Standalone: `.compass/Research/FEEDBACK-{slug}-{date}.md` ```bash compass-cli index add "" "research" 2>/dev/null || true ``` ## Save session `$PROJECT_ROOT/.compass/.state/sessions/-feedback-/transcript.md` ## Edge cases - **Feedback is a single sentence**: theme it as-is, note in the report that sample size is very small. - **All feedback falls into one theme**: report that accurately; don't force artificial categories. - **Contradictory feedback** (some users love X, others hate X): create a "Polarizing" sub-theme under the relevant category; do not average out the conflict. - **PO pastes in Vietnamese feedback**: if `lang=vi` or `spec_lang=vi`, keep quotes in Vietnamese in the report verbatim. - **Jira connected but returns 0 results with current filter**: warn the user, offer alternative filter options via AskUserQuestion. - **Feedback contains PII** (names, emails): redact before saving — replace with `[User A]`, `[User B]` etc. - **>50 signals**: auto-cap displayed quotes at 2 per theme, note total count in the category header. --- ## Final — Hand-off Print this closing message (AI translates per `$LANG` — see ux-rules Language Policy): `✓ Feedback themed. Next: `/compass:prioritize` to score top themes, or `/compass:research` to dig deeper into one.` Then stop. Do NOT auto-invoke the next workflow.