code-ai-installer 4.0.0 → 4.0.1-a

package/domains/product/locales/en/.agents/skills/user-story/SKILL.md

@@ -1,385 +1,401 @@
----
-name: user-story
-description: Writing user stories per INVEST — As a / I want / So that, separation of buyer/end-user
----
-# User Story
-
-> **Category:** Specification  ·  **Slug:** `user-story`
-
-## When to Use
-
-- In a PRD — every user-facing functionality.
-- In backlog grooming — transforming ideas into work items.
-- During epic breakdown — story-level.
-- In B2B — separation of buyer / end-user / admin stories.
-
-## Input
-
-| Field | Required | Description |
-|-------|:--------:|-------------|
-| User role | ✅ | Buyer / end-user / admin / champion |
-| JTBD / problem | ✅ | What we're solving |
-| Context | ✅ | When it occurs |
-
-## Data Sources
-
-1. `$jtbd-canvas` — jobs as starting point.
-2. `$problem-statement` — problem as «so that».
-3. Customer interviews — language of user.
-
-### Related Skills
-
-| Skill | What we take | When to invoke |
-|-------|-------------|----------------|
-| `jtbd-canvas` | Jobs → «I want» | Before stories |
-| `problem-statement` | Pain → «so that» | For justification |
-| `acceptance-criteria` | Contract | Per each story |
-| `epic-breakdown` | Grouping | After stories |
-
-## INVEST Criteria
-
-A good user story is:
-- **I**ndependent — can be developed separately
-- **N**egotiable — details can be discussed, not locked
-- **V**aluable — delivers customer value
-- **E**stimable — team can estimate effort
-- **S**mall — fits in 1-2 sprints (< 8 points)
-- **T**estable — clear pass/fail criteria
-
-## Format
-
-**Canonical:**
-> **As a** [role]  
-> **I want to** [action / capability]  
-> **So that** [benefit / outcome]
-
-Example:
-> **As a** DevOps engineer at mid-market SaaS  
-> **I want to** set team-level role templates in 2-3 clicks  
-> **So that** onboarding new teammates takes 10 min, not 2 hours
-
-## Protocol
-
-### Step 0 — Role Clarity (B2B Critical)
-
-**Different stories for different roles** — don't mix in one:
-- **End-user story** — daily use experience
-- **Buyer story** — purchase / evaluation decision
-- **Admin story** — configuration / governance
-- **Champion story** — advocacy / reporting
-
-In B2B — separate stories per role — different AC, different flows.
-
-### Step 1 — «So That» First
-
-Write backwards — benefit first, then action.
-
-Capture customer outcome:
-- ❌ «So that users can use the new feature» (circular)
-- ✅ «So that I can close the quarterly compliance audit in 2 hours instead of 2 days»
-
-Benefit — **specific, measurable**, tied to JTBD.
-
-### Step 2 — «I Want»
-
-Action or capability — what the user specifically does.
-
-- Atomic (one action)
-- Technology-agnostic («view team activity» not «click button on dashboard»)
-- User's language, not engineering
-
-### Step 3 — «As A»
-
-Specific role, not generic «user»:
-- «DevOps engineer at mid-market SaaS» > «user»
-- «Customer Success Manager with >100 accounts» > «user»
-
-Tie role to segment (if segment-specific feature).
-
-### Step 4 — INVEST Check
-
-Verify each story:
-- **I** — Can it be developed independently? If other stories are needed first → split.
-- **N** — Are details open for discussion? (Not over-specified)
-- **V** — Does customer get value? If only technical improvement → reformulate through user outcome.
-- **E** — Can team estimate it? If not — unknowns need spike.
-- **S** — Less than 2 sprints? If not — split.
-- **T** — Can clear pass/fail AC be written?
-
-### Step 5 — Splitting Big Stories
-
-Split techniques:
-- **Workflow:** «create → update → delete» → 3 stories
-- **Data:** different entity types (user → team → organization)
-- **Happy path + edges:** primary flow → error states as separate stories
-- **Acceptance criteria:** pull one scenario after another from AC
-
-### Step 6 — Story Sizing
-
-T-shirt or Story Points:
-- **S** (1-2 pts): < 1 day
-- **M** (3-5 pts): 1-3 days
-- **L** (8 pts): 4-7 days
-- **XL** (13+ pts): **split**
-
-### Step 7 — Negotiation Notes
-
-Add «Notes» field for:
-- Open questions
-- Assumptions
-- Suggested scope boundaries
-
-This signals to engineers: where negotiation room exists.
-
-### Step 8 — Add AC
-
-Each story requires AC via `$acceptance-criteria`.
-
-Without AC a story — is not ready for sprint.
-
-## Validation (Quality Gate)
-
-- [ ] Format: As a / I want / So that
-- [ ] Role — specific, not generic «user»
-- [ ] B2B: separated per role (if multi-role flow)
-- [ ] «So that» — specific, measurable benefit
-- [ ] INVEST passed (all 6)
-- [ ] Size ≤ 8 pts (or split)
-- [ ] AC attached
-- [ ] Notes / assumptions explicit
-
-## Handoff
-
-The result is input for:
-- **`acceptance-criteria`** — each story
-- **`epic-breakdown`** — grouping
-- **Engineering** → implementation
-- **QA** → test cases
-
-Format: story card (title + as-I-so-that + notes + AC link). Via `$handoff`.
-
-## Anti-patterns
-
-| Error | Why it's bad | How to do it right |
-|-------|-------------|-------------------|
-| «User» role | Lost context | Specific role/segment |
-| Vague «so that» | Lost value | Measurable benefit |
-| Implementation in story | Over-specified | Action, not UI |
-| Mix roles in one story | Confused AC | Separate per role |
-| Big (>13 pt) | Not estimable | Split using techniques |
-| No AC | No contract | Must have AC |
-| «As developer I want refactor» | Not customer value | Translate to user outcome |
-
-## Template
-
-```markdown
-## Story: [Short Title]
-
-**As a** [specific role]
-**I want to** [action]
-**So that** [measurable benefit]
-
-### Notes
-- Assumption: [X]
-- Open question: [Y]
-
-### Acceptance Criteria
-[via $acceptance-criteria]
-
-### Size
-M (3-5 points)
-
-### Dependencies
-- Story [#123] (prerequisite)
-```
-
-## Worked Example — TeamFlow User Stories for AI Summarization
-
-**Context:** 14 total user stories in PRD. Below — 7 key stories with proper INVEST + B2B role separation. Show splitting technique: end-user vs admin vs buyer.
-
-### Stories grouped by Role
-
----
-
-#### Role: People Manager (end-user) — Primary
-
-**Story S1: Enable AI Summarization for Upcoming 1:1**
-
-**As a** People manager with 5+ direct reports at mid-market B2B SaaS company on Team Tier  
-**I want to** toggle «AI assist» on for an upcoming 1:1 in one click  
-**So that** I can focus fully on conversation without worrying about notes
-
-**Notes:**
-- Default state: turned-off (opt-in respects privacy), user activates per-meeting
-- Admin can default-on for all new 1:1s (org setting, S7)
-- Assumption: managers prefer per-meeting control. Validated in Discovery E2-E4.
-- Open Q: does toggle apply to entire meeting or before/after toggle events? → clarify in AC
-
-**AC:** [see `$acceptance-criteria` S1 scenarios — 4 total]
-
-**Size:** S (2 points) — UI toggle, feature flag check, settings API  
-**Dependencies:** T1 (feature flag infrastructure)  
-**Role validation:** INVEST passes (independent, negotiable, valuable, estimable, small, testable)
-
----
-
-**Story S2: Receive AI Summary Within 60 Seconds of Meeting End**
-
-**As a** People manager who completed AI-enabled 1:1  
-**I want to** receive structured summary (topics, decisions, action items) within 60 seconds of meeting end  
-**So that** I can review and approve before moving to next task
-
-**Notes:**
-- Streaming UX acceptable — summary populates progressively within 60s
-- Quality threshold: ≥70% of summaries acceptable without major edits (measured via S3)
-- Assumption: 60s feels instant; 90s feels laggy. Validated via beta feedback in Wizard-of-Oz test.
-
-**AC:** [4 scenarios: happy path, streaming fallback, LLM timeout, meeting <5min skipped]
-
-**Size:** L (8 points) — core AI pipeline  
-**Dependencies:** S1 (must be enabled), T2 (LLM API integration)
-
----
-
-**Story S3: Review and Edit AI Summary Inline**
-
-**As a** manager reviewing AI-generated summary  
-**I want to** edit content inline (add, remove, modify) and mark approved  
-**So that** corrections become authoritative record, and future AI learns my style
-
-**Notes:**
-- Edit tracked separately from original AI version (both retained for audit + learning)
-- Approval state: «draft (AI)», «edited», «approved» (manager-confirmed)
-- Open Q: auto-approve after 7 days without edit? → default: no, keeps «awaiting review» status
-
-**AC:** [3 scenarios: basic edit, cancel edit, approved → read-only]
-
-**Size:** M (5 points) — rich text editor integration + state machine  
-**Dependencies:** S2 (must have summary first)
-
----
-
-**Story S4: Auto-Track Action Items Extracted from Summary**
-
-**As a** manager with approved AI summary  
-**I want** extracted action items automatically added to my follow-up queue with owner + due date (if LLM inferred)  
-**So that** I don't lose commitments between meetings
-
-**Notes:**
-- Extraction confidence threshold: >70% — low confidence items shown separately for manager review
-- Owner default: inferred from context (if «Sarah, please review» → Sarah), else manager
-- Due date default: «by next 1:1» (7 days), manager can override
-
-**AC:** [4 scenarios: standard extraction, low confidence review, owner inference, custom due date]
-
-**Size:** M (5 points)  
-**Dependencies:** S2
-
----
-
-**Story S5: See Action Items Reminder Before Next 1:1**
-
-**As a** manager with open action items from previous 1:1 with a report  
-**I want** those items surfaced 30 min before upcoming 1:1 via in-product notification + email  
-**So that** follow-through is top-of-mind and conversation flows naturally
-
-**Notes:**
-- 30 min window configurable (15 / 30 / 60 / disabled) — user setting
-- Items marked complete in reminder UI sync to both manager's and report's views
-
-**AC:** [3 scenarios: reminder fires, customized window, reminder disabled per user setting]
-
-**Size:** S (3 points) — rule-based, uses existing notification stack  
-**Dependencies:** S4 (needs action items existing)
-
----
-
-#### Role: Admin / HR Business Partner
-
-**Story S6: Configure Org-Wide AI Summarization Policy**
-
-**As an** admin (HR Business Partner or People Ops lead) for TeamFlow account on Team or Enterprise tier  
-**I want to** set org-wide AI policy (enabled / disabled / default-on for new managers)  
-**So that** I can align TeamFlow AI usage with company data governance decisions
-
-**Notes:**
-- 3 states: «disabled org-wide» (AI not available for anyone), «enabled — per-manager opt-in» (default), «enabled — default on for new managers»
-- Policy change notifies all managers
-- Policy setting audit-logged
-
-**AC:** [4 scenarios: 3 state transitions + audit log verification]
-
-**Size:** S (3 points)  
-**Dependencies:** S1 (per-meeting toggle must exist)
-
----
-
-**Story S7: Export Audit Log of AI Operations**
-
-**As an** admin responsible for compliance (SOC 2, GDPR)  
-**I want to** export audit log of all AI operations (summaries generated, edited, approved) for given time period  
-**So that** I can demonstrate compliance during audit + investigate incidents
-
-**Notes:**
-- Log entry: timestamp, actor (manager ID), action (generate/edit/approve/delete), meeting ID, summary ID  
-- Export formats: CSV, JSON
-- Retention: 365 days minimum (legal baseline)
-
-**AC:** [3 scenarios: export happy path, date range filter, empty range]
-
-**Size:** S (3 points) — uses existing audit log infrastructure  
-**Dependencies:** S1, S2, S3 (all AI operations logged)
-
----
-
-#### Role: Buyer (VP HR / CPO) — pulled through admin experience
-
-**Story S8 (pulled into PRD Appendix — Theme 2 dashboard scope):**
-
-VP HR visibility / aggregate dashboard stories — moved to separate PRD (aggregate-dashboard-v1), but linked:
-- «As VP HR, I want to see AI summarization adoption across teams so I can allocate training»  
-- «As CPO, I want action items velocity as leading indicator of team health»
-
-*These stories live in Aggregate Dashboard PRD — cross-referenced here.*
-
----
-
-### Story Sizing Summary
-
-| Story | Size | Total |
-|-------|:----:|------|
-| S1 Enable toggle | S (2) | Small, quick |
-| S2 Generate summary | L (8) | THE big one — core AI |
-| S3 Edit inline | M (5) | Rich editor |
-| S4 Extract action items | M (5) | Post-processing |
-| S5 Reminder | S (3) | Rule-based |
-| S6 Org policy | S (3) | Admin UI |
-| S7 Audit export | S (3) | Export feature |
-| **Total (primary flow):** | | **29 points** |
-
-Balance distribution (1 L + 2 M + 4 S) — healthy MVP shape. No XLs = no un-sized risks.
-
-### Splitting Techniques Applied
-
-- **By role:** S1-S5 end-user (manager), S6-S7 admin, S8 deferred to separate PRD (buyer dashboard)
-- **By workflow step:** Prep (defer — Q3), During meeting (S1-S2), Review (S3), Follow-through (S4-S5)
-- **Simple → Complex:** S2 ships «basic summary»; future ships «summary + themes» separately
-- **Happy path → Edge cases:** S2 happy path above, error-state handling via AC (not separate stories)
-
-### Story Dependencies DAG
-
-```
-T1 (feature flag infra)
-    └─→ S1 (enable toggle)
-           └─→ S2 (generate summary) ← depends on T2 (LLM API)
-                  ├─→ S3 (edit inline)
-                  └─→ S4 (extract action items)
-                         └─→ S5 (reminder)
-
-S6 (org policy) ← depends on S1 (per-meeting must exist first)
-S7 (audit export) ← depends on S1, S2, S3 (need operations to log)
-```
-
-**Critical path (for MVP):** T1 → S1 → S2 → S3 → S4 → S5 = ~26 points ≈ 6 weeks at 4 points/week velocity.
-
-> **user-story lesson:** B2B **always** has multiple roles. Separating manager (end-user) from admin (HR Business Partner) from buyer (VP HR) — 3 distinct AC sets, 3 distinct «so that» phrases. Without separation «As a user» = vague. Stories S6-S7 admin-focused — critical for companies with governance requirements. Our initial 14 stories included only manager — Discovery B3 interview revealed admin governance need; added S6-S7 late, prevented post-launch «admin has no control» complaints.
+---
+name: user-story
+description: Writing user stories per INVEST — As a / I want / So that, separation of buyer/end-user
+type: triggered
+domain: product
+owners:
+  - pm
+gates:
+  - PM
+tech: []
+topic: []
+triggers:
+  - "user-story"
+  - "user story"
+  - "пользовательская история"
+  - "INVEST"
+related: []
+budget_lines: 401
+schema_version: 1
+---
+# User Story
+
+> **Category:** Specification  ·  **Slug:** `user-story`
+
+## When to Use
+
+- In a PRD — every user-facing functionality.
+- In backlog grooming — transforming ideas into work items.
+- During epic breakdown — story-level.
+- In B2B — separation of buyer / end-user / admin stories.
+
+## Input
+
+| Field | Required | Description |
+|-------|:--------:|-------------|
+| User role | ✅ | Buyer / end-user / admin / champion |
+| JTBD / problem | ✅ | What we're solving |
+| Context | ✅ | When it occurs |
+
+## Data Sources
+
+1. `$jtbd-canvas` — jobs as starting point.
+2. `$problem-statement` — problem as «so that».
+3. Customer interviews — language of user.
+
+### Related Skills
+
+| Skill | What we take | When to invoke |
+|-------|-------------|----------------|
+| `jtbd-canvas` | Jobs → «I want» | Before stories |
+| `problem-statement` | Pain → «so that» | For justification |
+| `acceptance-criteria` | Contract | Per each story |
+| `epic-breakdown` | Grouping | After stories |
+
+## INVEST Criteria
+
+A good user story is:
+- **I**ndependent — can be developed separately
+- **N**egotiable — details can be discussed, not locked
+- **V**aluable — delivers customer value
+- **E**stimable — team can estimate effort
+- **S**mall — fits in 1-2 sprints (< 8 points)
+- **T**estable — clear pass/fail criteria
+
+## Format
+
+**Canonical:**
+> **As a** [role]  
+> **I want to** [action / capability]  
+> **So that** [benefit / outcome]
+
+Example:
+> **As a** DevOps engineer at mid-market SaaS  
+> **I want to** set team-level role templates in 2-3 clicks  
+> **So that** onboarding new teammates takes 10 min, not 2 hours
+
+## Protocol
+
+### Step 0 — Role Clarity (B2B Critical)
+
+**Different stories for different roles** — don't mix in one:
+- **End-user story** — daily use experience
+- **Buyer story** — purchase / evaluation decision
+- **Admin story** — configuration / governance
+- **Champion story** — advocacy / reporting
+
+In B2B — separate stories per role — different AC, different flows.
+
+### Step 1 — «So That» First
+
+Write backwards — benefit first, then action.
+
+Capture customer outcome:
+- ❌ «So that users can use the new feature» (circular)
+- ✅ «So that I can close the quarterly compliance audit in 2 hours instead of 2 days»
+
+Benefit — **specific, measurable**, tied to JTBD.
+
+### Step 2 — «I Want»
+
+Action or capability — what the user specifically does.
+
+- Atomic (one action)
+- Technology-agnostic («view team activity» not «click button on dashboard»)
+- User's language, not engineering
+
+### Step 3 — «As A»
+
+Specific role, not generic «user»:
+- «DevOps engineer at mid-market SaaS» > «user»
+- «Customer Success Manager with >100 accounts» > «user»
+
+Tie role to segment (if segment-specific feature).
+
+### Step 4 — INVEST Check
+
+Verify each story:
+- **I** — Can it be developed independently? If other stories are needed first → split.
+- **N** — Are details open for discussion? (Not over-specified)
+- **V** — Does customer get value? If only technical improvement → reformulate through user outcome.
+- **E** — Can team estimate it? If not — unknowns need spike.
+- **S** — Less than 2 sprints? If not — split.
+- **T** — Can clear pass/fail AC be written?
+
+### Step 5 — Splitting Big Stories
+
+Split techniques:
+- **Workflow:** «create → update → delete» → 3 stories
+- **Data:** different entity types (user → team → organization)
+- **Happy path + edges:** primary flow → error states as separate stories
+- **Acceptance criteria:** pull one scenario after another from AC
+
+### Step 6 — Story Sizing
+
+T-shirt or Story Points:
+- **S** (1-2 pts): < 1 day
+- **M** (3-5 pts): 1-3 days
+- **L** (8 pts): 4-7 days
+- **XL** (13+ pts): **split**
+
+### Step 7 — Negotiation Notes
+
+Add «Notes» field for:
+- Open questions
+- Assumptions
+- Suggested scope boundaries
+
+This signals to engineers: where negotiation room exists.
+
+### Step 8 — Add AC
+
+Each story requires AC via `$acceptance-criteria`.
+
+Without AC a story — is not ready for sprint.
+
+## Validation (Quality Gate)
+
+- [ ] Format: As a / I want / So that
+- [ ] Role — specific, not generic «user»
+- [ ] B2B: separated per role (if multi-role flow)
+- [ ] «So that» — specific, measurable benefit
+- [ ] INVEST passed (all 6)
+- [ ] Size ≤ 8 pts (or split)
+- [ ] AC attached
+- [ ] Notes / assumptions explicit
+
+## Handoff
+
+The result is input for:
+- **`acceptance-criteria`** — each story
+- **`epic-breakdown`** — grouping
+- **Engineering** → implementation
+- **QA** → test cases
+
+Format: story card (title + as-I-so-that + notes + AC link). Via `$handoff`.
+
+## Anti-patterns
+
+| Error | Why it's bad | How to do it right |
+|-------|-------------|-------------------|
+| «User» role | Lost context | Specific role/segment |
+| Vague «so that» | Lost value | Measurable benefit |
+| Implementation in story | Over-specified | Action, not UI |
+| Mix roles in one story | Confused AC | Separate per role |
+| Big (>13 pt) | Not estimable | Split using techniques |
+| No AC | No contract | Must have AC |
+| «As developer I want refactor» | Not customer value | Translate to user outcome |
+
+## Template
+
+```markdown
+## Story: [Short Title]
+
+**As a** [specific role]
+**I want to** [action]
+**So that** [measurable benefit]
+
+### Notes
+- Assumption: [X]
+- Open question: [Y]
+
+### Acceptance Criteria
+[via $acceptance-criteria]
+
+### Size
+M (3-5 points)
+
+### Dependencies
+- Story [#123] (prerequisite)
+```
+
+## Worked Example — TeamFlow User Stories for AI Summarization
+
+**Context:** 14 total user stories in PRD. Below — 7 key stories with proper INVEST + B2B role separation. Show splitting technique: end-user vs admin vs buyer.
+
+### Stories grouped by Role
+
+---
+
+#### Role: People Manager (end-user) — Primary
+
+**Story S1: Enable AI Summarization for Upcoming 1:1**
+
+**As a** People manager with 5+ direct reports at mid-market B2B SaaS company on Team Tier  
+**I want to** toggle «AI assist» on for an upcoming 1:1 in one click  
+**So that** I can focus fully on conversation without worrying about notes
+
+**Notes:**
+- Default state: turned-off (opt-in respects privacy), user activates per-meeting
+- Admin can default-on for all new 1:1s (org setting, S7)
+- Assumption: managers prefer per-meeting control. Validated in Discovery E2-E4.
+- Open Q: does toggle apply to entire meeting or before/after toggle events? → clarify in AC
+
+**AC:** [see `$acceptance-criteria` S1 scenarios — 4 total]
+
+**Size:** S (2 points) — UI toggle, feature flag check, settings API  
+**Dependencies:** T1 (feature flag infrastructure)  
+**Role validation:** INVEST passes (independent, negotiable, valuable, estimable, small, testable)
+
+---
+
+**Story S2: Receive AI Summary Within 60 Seconds of Meeting End**
+
+**As a** People manager who completed AI-enabled 1:1  
+**I want to** receive structured summary (topics, decisions, action items) within 60 seconds of meeting end  
+**So that** I can review and approve before moving to next task
+
+**Notes:**
+- Streaming UX acceptable — summary populates progressively within 60s
+- Quality threshold: ≥70% of summaries acceptable without major edits (measured via S3)
+- Assumption: 60s feels instant; 90s feels laggy. Validated via beta feedback in Wizard-of-Oz test.
+
+**AC:** [4 scenarios: happy path, streaming fallback, LLM timeout, meeting <5min skipped]
+
+**Size:** L (8 points) — core AI pipeline  
+**Dependencies:** S1 (must be enabled), T2 (LLM API integration)
+
+---
+
+**Story S3: Review and Edit AI Summary Inline**
+
+**As a** manager reviewing AI-generated summary  
+**I want to** edit content inline (add, remove, modify) and mark approved  
+**So that** corrections become authoritative record, and future AI learns my style
+
+**Notes:**
+- Edit tracked separately from original AI version (both retained for audit + learning)
+- Approval state: «draft (AI)», «edited», «approved» (manager-confirmed)
+- Open Q: auto-approve after 7 days without edit? → default: no, keeps «awaiting review» status
+
+**AC:** [3 scenarios: basic edit, cancel edit, approved → read-only]
+
+**Size:** M (5 points) — rich text editor integration + state machine  
+**Dependencies:** S2 (must have summary first)
+
+---
+
+**Story S4: Auto-Track Action Items Extracted from Summary**
+
+**As a** manager with approved AI summary  
+**I want** extracted action items automatically added to my follow-up queue with owner + due date (if LLM inferred)  
+**So that** I don't lose commitments between meetings
+
+**Notes:**
+- Extraction confidence threshold: >70% — low confidence items shown separately for manager review
+- Owner default: inferred from context (if «Sarah, please review» → Sarah), else manager
+- Due date default: «by next 1:1» (7 days), manager can override
+
+**AC:** [4 scenarios: standard extraction, low confidence review, owner inference, custom due date]
+
+**Size:** M (5 points)  
+**Dependencies:** S2
+
+---
+
+**Story S5: See Action Items Reminder Before Next 1:1**
+
+**As a** manager with open action items from previous 1:1 with a report  
+**I want** those items surfaced 30 min before upcoming 1:1 via in-product notification + email  
+**So that** follow-through is top-of-mind and conversation flows naturally
+
+**Notes:**
+- 30 min window configurable (15 / 30 / 60 / disabled) — user setting
+- Items marked complete in reminder UI sync to both manager's and report's views
+
+**AC:** [3 scenarios: reminder fires, customized window, reminder disabled per user setting]
+
+**Size:** S (3 points) — rule-based, uses existing notification stack  
+**Dependencies:** S4 (needs action items existing)
+
+---
+
+#### Role: Admin / HR Business Partner
+
+**Story S6: Configure Org-Wide AI Summarization Policy**
+
+**As an** admin (HR Business Partner or People Ops lead) for TeamFlow account on Team or Enterprise tier  
+**I want to** set org-wide AI policy (enabled / disabled / default-on for new managers)  
+**So that** I can align TeamFlow AI usage with company data governance decisions
+
+**Notes:**
+- 3 states: «disabled org-wide» (AI not available for anyone), «enabled — per-manager opt-in» (default), «enabled — default on for new managers»
+- Policy change notifies all managers
+- Policy setting audit-logged
+
+**AC:** [4 scenarios: 3 state transitions + audit log verification]
+
+**Size:** S (3 points)  
+**Dependencies:** S1 (per-meeting toggle must exist)
+
+---
+
+**Story S7: Export Audit Log of AI Operations**
+
+**As an** admin responsible for compliance (SOC 2, GDPR)  
+**I want to** export audit log of all AI operations (summaries generated, edited, approved) for given time period  
+**So that** I can demonstrate compliance during audit + investigate incidents
+
+**Notes:**
+- Log entry: timestamp, actor (manager ID), action (generate/edit/approve/delete), meeting ID, summary ID  
+- Export formats: CSV, JSON
+- Retention: 365 days minimum (legal baseline)
+
+**AC:** [3 scenarios: export happy path, date range filter, empty range]
+
+**Size:** S (3 points) — uses existing audit log infrastructure  
+**Dependencies:** S1, S2, S3 (all AI operations logged)
+
+---
+
+#### Role: Buyer (VP HR / CPO) — pulled through admin experience
+
+**Story S8 (pulled into PRD Appendix — Theme 2 dashboard scope):**
+
+VP HR visibility / aggregate dashboard stories — moved to separate PRD (aggregate-dashboard-v1), but linked:
+- «As VP HR, I want to see AI summarization adoption across teams so I can allocate training»  
+- «As CPO, I want action items velocity as leading indicator of team health»
+
+*These stories live in Aggregate Dashboard PRD — cross-referenced here.*
+
+---
+
+### Story Sizing Summary
+
+| Story | Size | Total |
+|-------|:----:|------|
+| S1 Enable toggle | S (2) | Small, quick |
+| S2 Generate summary | L (8) | THE big one — core AI |
+| S3 Edit inline | M (5) | Rich editor |
+| S4 Extract action items | M (5) | Post-processing |
+| S5 Reminder | S (3) | Rule-based |
+| S6 Org policy | S (3) | Admin UI |
+| S7 Audit export | S (3) | Export feature |
+| **Total (primary flow):** | | **29 points** |
+
+Balance distribution (1 L + 2 M + 4 S) — healthy MVP shape. No XLs = no un-sized risks.
+
+### Splitting Techniques Applied
+
+- **By role:** S1-S5 end-user (manager), S6-S7 admin, S8 deferred to separate PRD (buyer dashboard)
+- **By workflow step:** Prep (defer — Q3), During meeting (S1-S2), Review (S3), Follow-through (S4-S5)
+- **Simple → Complex:** S2 ships «basic summary»; future ships «summary + themes» separately
+- **Happy path → Edge cases:** S2 happy path above, error-state handling via AC (not separate stories)
+
+### Story Dependencies DAG
+
+```
+T1 (feature flag infra)
+    └─→ S1 (enable toggle)
+           └─→ S2 (generate summary) ← depends on T2 (LLM API)
+                  ├─→ S3 (edit inline)
+                  └─→ S4 (extract action items)
+                         └─→ S5 (reminder)
+
+S6 (org policy) ← depends on S1 (per-meeting must exist first)
+S7 (audit export) ← depends on S1, S2, S3 (need operations to log)
+```
+
+**Critical path (for MVP):** T1 → S1 → S2 → S3 → S4 → S5 = ~26 points ≈ 6 weeks at 4 points/week velocity.
+
+> **user-story lesson:** B2B **always** has multiple roles. Separating manager (end-user) from admin (HR Business Partner) from buyer (VP HR) — 3 distinct AC sets, 3 distinct «so that» phrases. Without separation «As a user» = vague. Stories S6-S7 admin-focused — critical for companies with governance requirements. Our initial 14 stories included only manager — Discovery B3 interview revealed admin governance need; added S6-S7 late, prevented post-launch «admin has no control» complaints.