code-ai-installer 4.0.0 → 4.0.1-b

package/domains/product/locales/en/.agents/skills/prd-template/SKILL.md

@@ -1,6 +1,23 @@
 ---
 name: prd-template
 description: Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout
+type: triggered
+domain: product
+owners:
+  - pm
+gates:
+  - PM
+tech: []
+topic: []
+triggers:
+  - "prd-template"
+  - "PRD"
+  - "product requirements document"
+  - "спецификация продукта"
+related:
+  - prd-template-reference
+budget_lines: 250
+schema_version: 1
 ---
 # PRD Template (B2B SaaS)
 
@@ -230,261 +247,4 @@ Format: PRD markdown (can be converted to PDF). Via `$handoff`.
 
 ## Template
 
-```markdown
-# PRD: [Initiative Name]
-
-**Owner:** [PM]  ·  **Status:** Draft  ·  **Target:** Q2 2026
-
-## TL;DR
-[3-5 sentences]
-
-## Problem
-[via $problem-statement]
-
-## Goals
-1. ...
-
-## Non-Goals (out of scope)
-- ...
-
-## Strategy Fit
-- OKR: [link]
-- NSM impact: [expected]
-
-## Solution
-[High-level approach]
-
-## User Stories
-### Story 1: [title]
-[via $user-story]
-
-**Acceptance Criteria:** [via $acceptance-criteria]
-
-## Non-Functional Requirements
-- Performance: p95 < 200ms
-- Availability: 99.9%
-- Security: SSO, audit logs
-- Compliance: SOC 2 Type II
-...
-
-## Success Criteria
-| Metric | Baseline | 30d | 90d | 6mo |
-| Primary | | | | |
-
-## Risks
-...
-
-## Rollout
-1. Dogfood week N
-2. Beta N customers week N+1
-3. GA week N+2
-4. Rollback: feature flag
-
-## Dependencies
-- Platform: [what needed]
-- Security: [review timeline]
-
-## Open Questions
-- [Q, owner, by when]
-```
-
-## Worked Example — TeamFlow PRD: AI 1:1 Summarization MVP (excerpt)
-
-```markdown
-# PRD: AI 1:1 Summarization (Team Tier MVP)
-
-**Owner:** Alex K., PM  ·  **Status:** Approved (post-Mediator synthesis)  ·  **Target Release:** Q2 2026 (GA target June 30)
-**Contributors:** Eng Lead (Priya S.), Design Lead (Jordan M.), Data Lead (Sam P.), Legal (Mia R.)
-**Links:** [OKR Q2 Product doc] · [Figma designs] · [Discovery Brief session-1-handoff.md] · [Mediator synthesis session-4-handoff.md]
-
-## TL;DR
-AI Summarization turns 1:1s from a 3-hour/week admin burden into a 30-minute confident workflow. Managers speak, TeamFlow captures, summarizes, extracts action items, reminds. Launched under the Team Tier (+$8/seat). Success: 40 account upgrades + 65% manager adoption in AI-tier accounts within the first 90 days.
-
-## Problem
-People managers (5-15 direct reports) at mid-market B2B SaaS companies (100-1000 employees) experience 45-60 min/week per report on 1:1 prep + note-taking + follow-up (3-4 hrs/week for a manager with 8 reports), because TeamFlow provides structured templates but no capture or synthesis capability.
-
-**Evidence:** 6 of 8 customer interviews + 120 support tickets Q1 2026 + internal analytics. Impact: ~3,000 affected managers across 200-customer base. Business: $3-5M ARR expansion opportunity + 10pp NRR lift potential through Team Tier upgrade path.
-
-*Full analysis: [problem-statement #1 in Discovery Brief]*
-
-## Goals
-1. **Adoption:** 65% of managers in AI-tier accounts use summarization weekly by Day 90
-2. **Tier conversion:** 40 accounts upgrade from Core to Team tier (+$8/seat) by Q2 end
-3. **Action items lift:** 60% → 75% completion rate in AI-tier accounts
-4. **Quality:** Summary acceptance rate (not edited >50%) ≥ 70%
-
-## Non-Goals (explicitly out of scope)
-- Real-time transcription during 1:1 (Q3 target — per Mediator scope decision)
-- Pre-1:1 prep card generation (Q3 — requires multi-meeting data)
-- Group meeting / team meeting summarization (future, scope creep control)
-- Non-English language support (English-first, EU languages Q4)
-- Mobile-native experience (web responsive only for MVP)
-- Integration with external meeting tools (Zoom/Meet) — Q3-Q4
-
-## Strategy Fit
-- **OKR:** Drives O1.1 (40 tier upgrades), O1.2 (65% manager adoption), O1.3 (NPS ≥50), O1.4 (action items 60→75%)
-- **NSM:** Primary driver of «Sticky Manager Ratio» — target 20% by Day 90
-- **Roadmap:** Theme 1 «AI-Assisted 1:1 Workflow» — Now bucket item
-
-## Solution (High-Level)
-
-Manager workflow:
-1. Before 1:1: Toggle «AI assist» on (or default-on for AI tier accounts)
-2. During 1:1: Normal conversation in TeamFlow 1:1 interface. No typing required (audio optional via browser recording with user consent).
-3. After 1:1: AI generates summary + extracts action items. Manager reviews, edits (inline), or approves.
-4. Before next 1:1: Action items from previous summary surface prominently.
-
-Architecture: summaries generated via OpenAI GPT-4 API (regional endpoints: US, EU) on opt-in basis. No data retention in LLM provider (zero-retention API tier). Summaries stored in TeamFlow DB with customer-owned encryption keys.
-
-**Alternatives considered:**
-- Fine-tuned on-premise LLM: rejected (quality gap, infra complexity, Q2 timeline)
-- Anthropic Claude via Bedrock: **accepted as fallback** (contract in place for redundancy)
-- Third-party (Otter.ai embed): rejected (privacy architecture not sufficient for HR data)
-
-## User Stories (top 5 from PRD backlog; full list: 14 stories)
-
-### Story S1: Enable AI Summarization for New 1:1
-**As a** People manager with 5+ reports at mid-market B2B SaaS company (Team Tier account)
-**I want to** enable AI summarization for upcoming 1:1 with one click
-**So that** I can focus fully on conversation without worrying about note-taking
-
-**AC:** [via `$acceptance-criteria` — 4 scenarios: happy path, already-enabled, admin-disabled-at-org-level, AI provider unavailable fallback]
-
-**Size:** S (2 points) · **Depends on:** T1 (feature flag infra)
-
-### Story S2: Generate Summary After 1:1 Ends
-**As a** manager who just finished an AI-enabled 1:1
-**I want to** receive summary + action items within 60 seconds of meeting end
-**So that** I can review and approve before moving to my next task
-
-**AC:** [4 scenarios incl. streaming delivery UX]
-**Size:** L (8 points) — THE core AI work · **Depends on:** S1
-
-### Story S3: Review and Edit Summary Inline
-**As a** manager reviewing AI-generated summary
-**I want to** edit content inline and mark it approved
-**So that** corrections become the authoritative record and AI learns preferences over time
-
-**AC:** [3 scenarios]
-**Size:** M (5 points)
-
-### Story S4: Auto-Extract and Track Action Items
-**As a** manager with approved AI summary
-**I want** action items automatically extracted and tracked in my 1:1 follow-up queue
-**So that** I don't lose commitments between meetings
-
-**AC:** [4 scenarios incl. detection confidence threshold]
-**Size:** M (5 points) · **Depends on:** S2
-
-### Story S5: Receive Action Items Reminder Before Next 1:1
-**As a** manager with open action items from previous 1:1
-**I want** them surfaced 30 min before next 1:1
-**So that** follow-through is top-of-mind
-
-**AC:** [3 scenarios]
-**Size:** S (3 points) — rule-based, not AI · **Depends on:** S4
-
-## Non-Functional Requirements
-
-- **Performance:** 
-  - Summary generation: p95 ≤ 60s from meeting end
-  - Summary load time: ≤ 500ms
-- **Availability:** 99.9% (excluding 3rd-party LLM outages, for which a documented fallback to secondary provider exists)
-- **Security:**
-  - All transcript data encrypted at rest (AES-256) and in transit (TLS 1.3)
-  - Per-customer encryption key (BYOK for enterprise)
-  - Zero data retention in LLM provider (use OpenAI/Anthropic zero-retention API tiers)
-  - Audit log of every AI operation (summary generated, edited, approved)
-- **Compliance:**
-  - SOC 2 Type II attestation extended to cover AI features (audit Q2)
-  - GDPR compliant — EU customer data stays in EU region (Azure West Europe)
-  - No training on customer data (contractual with providers)
-- **Scalability:** Support 5,000 summaries/day Q2, scale to 50,000/day by Q4
-- **Observability:** Full instrumentation — generation latency, quality rating, edit rate, action items extraction accuracy
-- **Privacy UX:**
-  - Opt-in per meeting (manager control)
-  - Org-level disable toggle (admin can turn off company-wide)
-  - Retention controls (30 / 90 / 365 / 730 days, customer choice)
-
-## Success Criteria
-
-| Metric | Baseline (pre-launch) | 30d target | 90d target | 6mo target |
-|--------|:--------------------:|:----------:|:----------:|:----------:|
-| **Sticky Manager Ratio (NSM)** | 0% | 10% | 20% | 35% |
-| Team Tier upgrades | 0 accounts | 15 accounts | 40 accounts | 80 accounts |
-| Action items completion (AI-tier) | 60% baseline | 68% | 75% | 80% |
-| AI summary acceptance (no major edits >50%) | N/A | 65% | 70% | 75% |
-| NPS in AI tier | 45 (blended TeamFlow NPS) | 48 | 50 | 55 |
-| **Guardrails:** | | | | |
-| Gross churn (overall) | 8% annual | ≤8% | ≤8% | ≤8% |
-| Support ticket volume / AI issues | N/A | <5% of total | <3% | <2% |
-| p95 latency | N/A | <60s | <45s | <30s |
-
-## Risks & Mitigations (Top 5)
-
-| # | Risk | Prob | Impact | Mitigation |
-|---|------|:----:|:------:|------------|
-| 1 | LLM quality below 85% acceptable — user trust collapses | M | H | Wizard-of-Oz validated 78% pre-launch; human-review layer feature flag ready; secondary LLM provider contract |
-| 2 | Privacy concern backlash (surveillance perception) | M | H | Opt-in per meeting; transparent data model; communicated early to customer advisory; legal review completed |
-| 3 | SOC 2 Type II audit delays block enterprise deals | M | H | Parallel-track audit started Week 1; conditional deal commitments acceptable; fallback Type II deferred to Q3 |
-| 4 | OpenAI pricing / terms change mid-quarter | L | M | Anthropic contract in place; API abstraction layer allows swap in <1 week |
-| 5 | Action items extraction accuracy low → user distrust | M | M | Confidence threshold surfacing; manager always can edit; iterative prompt tuning |
-
-## Rollout Plan
-
-**Phase 1 — Internal Dogfood (Week 6-7):**
-- TeamFlow employees use feature internally
-- Top 5 issues identified + resolved pre-beta
-
-**Phase 2 — Beta (Week 8-10):**
-- 10 Design Partner customers (curated mix: SMB 4, mid-market 4, enterprise 2)
-- Weekly customer interviews + survey
-- NPS target ≥45 during beta
-
-**Phase 3 — Limited GA (Week 11-12):**
-- Feature flag exposed to 25% of AI-tier-upgraded customers
-- Monitor guardrails; gate full rollout on green metrics
-
-**Phase 4 — Full GA (Week 13+):**
-- Feature flag to 100% of AI tier
-- Sales and CS enabled; marketing campaign triggered
-- Tier upgrade funnel open for non-AI-tier customers
-
-**Rollback triggers:**
-- Error rate >2% for >15 minutes → feature flag disable
-- p95 latency >2 min for >30 minutes → disable
-- Support ticket spike 5× baseline → investigation + possible disable
-- Critical privacy incident → immediate disable + customer notification
-
-## Dependencies
-
-- **Platform team:** Event instrumentation schema (Week 4 delivery) — blocker for Week 6 dogfood
-- **Security team:** SOC 2 Type II audit sign-off (Week 10) — blocker for Enterprise deals, not for GA
-- **Legal:** DPA updates sent to customers (Week 8)
-- **Sales:** Enablement training session (Week 11)
-- **CS:** Runbook + training (Week 10)
-- **Marketing:** Landing page + email campaign (Week 11)
-
-## Open Questions
-
-1. Pricing — Team Tier +$8/seat confirmed? (Owner: VP Sales, by Week 3)
-2. Enterprise customers — BYOK encryption delivery path? (Owner: Eng Lead, by Week 5)
-3. Audit log retention — default 365 days, customer override? (Owner: Legal, by Week 4)
-4. Beta customer selection finalized? (Owner: CS Lead, by Week 7)
-
----
-
-## Appendix A: Full User Stories List (14 stories)
-[Link to Stories spreadsheet]
-
-## Appendix B: Design References
-[Figma link]
-
-## Appendix C: Mediator Synthesis Rationale
-[Link to session-4-handoff.md]
-
-## Appendix D: Metric Plan Details
-[Link to metric-plan.md by Data Analyst]
-```
-
-> **PRD lesson:** every section traces back to Discovery / Strategy / Mediator artifacts — this prevents the "we discussed it somewhere" syndrome. NFR in B2B SaaS is critical — privacy + compliance take up more space than performance. Rollback triggers are numeric — not «if it breaks», but «error rate >2% for >15 min». Non-goals are explicit to prevent scope creep in implementation.
+> 📦 <!--SPLIT-POINTER--> Полный блок — в `$prd-template-reference` (§ Template).

package/domains/product/locales/en/.agents/skills/prd-template-reference/SKILL.md

@@ -0,0 +1,285 @@
+---
+name: prd-template-reference
+description: Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout — reference (code & worked-example library)
+type: reference
+domain: product
+owners:
+  - pm
+gates:
+  - PM
+tech: []
+topic: []
+triggers:
+  - "prd-template"
+  - "PRD"
+  - "product requirements document"
+  - "спецификация продукта"
+related:
+  - prd-template
+budget_lines: 285
+schema_version: 1
+---
+# prd-template — Reference (code & worked-example library)
+
+> Verbatim code blocks and the worked example extracted from `$prd-template` (workflow). Copy-paste library; the procedure lives in the workflow skill.
+
+## Template
+
+```markdown
+# PRD: [Initiative Name]
+
+**Owner:** [PM]  ·  **Status:** Draft  ·  **Target:** Q2 2026
+
+## TL;DR
+[3-5 sentences]
+
+## Problem
+[via $problem-statement]
+
+## Goals
+1. ...
+
+## Non-Goals (out of scope)
+- ...
+
+## Strategy Fit
+- OKR: [link]
+- NSM impact: [expected]
+
+## Solution
+[High-level approach]
+
+## User Stories
+### Story 1: [title]
+[via $user-story]
+
+**Acceptance Criteria:** [via $acceptance-criteria]
+
+## Non-Functional Requirements
+- Performance: p95 < 200ms
+- Availability: 99.9%
+- Security: SSO, audit logs
+- Compliance: SOC 2 Type II
+...
+
+## Success Criteria
+| Metric | Baseline | 30d | 90d | 6mo |
+| Primary | | | | |
+
+## Risks
+...
+
+## Rollout
+1. Dogfood week N
+2. Beta N customers week N+1
+3. GA week N+2
+4. Rollback: feature flag
+
+## Dependencies
+- Platform: [what needed]
+- Security: [review timeline]
+
+## Open Questions
+- [Q, owner, by when]
+```
+
+## Worked Example — TeamFlow PRD: AI 1:1 Summarization MVP (excerpt)
+
+```markdown
+# PRD: AI 1:1 Summarization (Team Tier MVP)
+
+**Owner:** Alex K., PM  ·  **Status:** Approved (post-Mediator synthesis)  ·  **Target Release:** Q2 2026 (GA target June 30)
+**Contributors:** Eng Lead (Priya S.), Design Lead (Jordan M.), Data Lead (Sam P.), Legal (Mia R.)
+**Links:** [OKR Q2 Product doc] · [Figma designs] · [Discovery Brief session-1-handoff.md] · [Mediator synthesis session-4-handoff.md]
+
+## TL;DR
+AI Summarization turns 1:1s from a 3-hour/week admin burden into a 30-minute confident workflow. Managers speak, TeamFlow captures, summarizes, extracts action items, reminds. Launched under the Team Tier (+$8/seat). Success: 40 account upgrades + 65% manager adoption in AI-tier accounts within the first 90 days.
+
+## Problem
+People managers (5-15 direct reports) at mid-market B2B SaaS companies (100-1000 employees) experience 45-60 min/week per report on 1:1 prep + note-taking + follow-up (3-4 hrs/week for a manager with 8 reports), because TeamFlow provides structured templates but no capture or synthesis capability.
+
+**Evidence:** 6 of 8 customer interviews + 120 support tickets Q1 2026 + internal analytics. Impact: ~3,000 affected managers across 200-customer base. Business: $3-5M ARR expansion opportunity + 10pp NRR lift potential through Team Tier upgrade path.
+
+*Full analysis: [problem-statement #1 in Discovery Brief]*
+
+## Goals
+1. **Adoption:** 65% of managers in AI-tier accounts use summarization weekly by Day 90
+2. **Tier conversion:** 40 accounts upgrade from Core to Team tier (+$8/seat) by Q2 end
+3. **Action items lift:** 60% → 75% completion rate in AI-tier accounts
+4. **Quality:** Summary acceptance rate (not edited >50%) ≥ 70%
+
+## Non-Goals (explicitly out of scope)
+- Real-time transcription during 1:1 (Q3 target — per Mediator scope decision)
+- Pre-1:1 prep card generation (Q3 — requires multi-meeting data)
+- Group meeting / team meeting summarization (future, scope creep control)
+- Non-English language support (English-first, EU languages Q4)
+- Mobile-native experience (web responsive only for MVP)
+- Integration with external meeting tools (Zoom/Meet) — Q3-Q4
+
+## Strategy Fit
+- **OKR:** Drives O1.1 (40 tier upgrades), O1.2 (65% manager adoption), O1.3 (NPS ≥50), O1.4 (action items 60→75%)
+- **NSM:** Primary driver of «Sticky Manager Ratio» — target 20% by Day 90
+- **Roadmap:** Theme 1 «AI-Assisted 1:1 Workflow» — Now bucket item
+
+## Solution (High-Level)
+
+Manager workflow:
+1. Before 1:1: Toggle «AI assist» on (or default-on for AI tier accounts)
+2. During 1:1: Normal conversation in TeamFlow 1:1 interface. No typing required (audio optional via browser recording with user consent).
+3. After 1:1: AI generates summary + extracts action items. Manager reviews, edits (inline), or approves.
+4. Before next 1:1: Action items from previous summary surface prominently.
+
+Architecture: summaries generated via OpenAI GPT-4 API (regional endpoints: US, EU) on opt-in basis. No data retention in LLM provider (zero-retention API tier). Summaries stored in TeamFlow DB with customer-owned encryption keys.
+
+**Alternatives considered:**
+- Fine-tuned on-premise LLM: rejected (quality gap, infra complexity, Q2 timeline)
+- Anthropic Claude via Bedrock: **accepted as fallback** (contract in place for redundancy)
+- Third-party (Otter.ai embed): rejected (privacy architecture not sufficient for HR data)
+
+## User Stories (top 5 from PRD backlog; full list: 14 stories)
+
+### Story S1: Enable AI Summarization for New 1:1
+**As a** People manager with 5+ reports at mid-market B2B SaaS company (Team Tier account)
+**I want to** enable AI summarization for upcoming 1:1 with one click
+**So that** I can focus fully on conversation without worrying about note-taking
+
+**AC:** [via `$acceptance-criteria` — 4 scenarios: happy path, already-enabled, admin-disabled-at-org-level, AI provider unavailable fallback]
+
+**Size:** S (2 points) · **Depends on:** T1 (feature flag infra)
+
+### Story S2: Generate Summary After 1:1 Ends
+**As a** manager who just finished an AI-enabled 1:1
+**I want to** receive summary + action items within 60 seconds of meeting end
+**So that** I can review and approve before moving to my next task
+
+**AC:** [4 scenarios incl. streaming delivery UX]
+**Size:** L (8 points) — THE core AI work · **Depends on:** S1
+
+### Story S3: Review and Edit Summary Inline
+**As a** manager reviewing AI-generated summary
+**I want to** edit content inline and mark it approved
+**So that** corrections become the authoritative record and AI learns preferences over time
+
+**AC:** [3 scenarios]
+**Size:** M (5 points)
+
+### Story S4: Auto-Extract and Track Action Items
+**As a** manager with approved AI summary
+**I want** action items automatically extracted and tracked in my 1:1 follow-up queue
+**So that** I don't lose commitments between meetings
+
+**AC:** [4 scenarios incl. detection confidence threshold]
+**Size:** M (5 points) · **Depends on:** S2
+
+### Story S5: Receive Action Items Reminder Before Next 1:1
+**As a** manager with open action items from previous 1:1
+**I want** them surfaced 30 min before next 1:1
+**So that** follow-through is top-of-mind
+
+**AC:** [3 scenarios]
+**Size:** S (3 points) — rule-based, not AI · **Depends on:** S4
+
+## Non-Functional Requirements
+
+- **Performance:** 
+  - Summary generation: p95 ≤ 60s from meeting end
+  - Summary load time: ≤ 500ms
+- **Availability:** 99.9% (excluding 3rd-party LLM outages, for which a documented fallback to secondary provider exists)
+- **Security:**
+  - All transcript data encrypted at rest (AES-256) and in transit (TLS 1.3)
+  - Per-customer encryption key (BYOK for enterprise)
+  - Zero data retention in LLM provider (use OpenAI/Anthropic zero-retention API tiers)
+  - Audit log of every AI operation (summary generated, edited, approved)
+- **Compliance:**
+  - SOC 2 Type II attestation extended to cover AI features (audit Q2)
+  - GDPR compliant — EU customer data stays in EU region (Azure West Europe)
+  - No training on customer data (contractual with providers)
+- **Scalability:** Support 5,000 summaries/day Q2, scale to 50,000/day by Q4
+- **Observability:** Full instrumentation — generation latency, quality rating, edit rate, action items extraction accuracy
+- **Privacy UX:**
+  - Opt-in per meeting (manager control)
+  - Org-level disable toggle (admin can turn off company-wide)
+  - Retention controls (30 / 90 / 365 / 730 days, customer choice)
+
+## Success Criteria
+
+| Metric | Baseline (pre-launch) | 30d target | 90d target | 6mo target |
+|--------|:--------------------:|:----------:|:----------:|:----------:|
+| **Sticky Manager Ratio (NSM)** | 0% | 10% | 20% | 35% |
+| Team Tier upgrades | 0 accounts | 15 accounts | 40 accounts | 80 accounts |
+| Action items completion (AI-tier) | 60% baseline | 68% | 75% | 80% |
+| AI summary acceptance (no major edits >50%) | N/A | 65% | 70% | 75% |
+| NPS in AI tier | 45 (blended TeamFlow NPS) | 48 | 50 | 55 |
+| **Guardrails:** | | | | |
+| Gross churn (overall) | 8% annual | ≤8% | ≤8% | ≤8% |
+| Support ticket volume / AI issues | N/A | <5% of total | <3% | <2% |
+| p95 latency | N/A | <60s | <45s | <30s |
+
+## Risks & Mitigations (Top 5)
+
+| # | Risk | Prob | Impact | Mitigation |
+|---|------|:----:|:------:|------------|
+| 1 | LLM quality below 85% acceptable — user trust collapses | M | H | Wizard-of-Oz validated 78% pre-launch; human-review layer feature flag ready; secondary LLM provider contract |
+| 2 | Privacy concern backlash (surveillance perception) | M | H | Opt-in per meeting; transparent data model; communicated early to customer advisory; legal review completed |
+| 3 | SOC 2 Type II audit delays block enterprise deals | M | H | Parallel-track audit started Week 1; conditional deal commitments acceptable; fallback Type II deferred to Q3 |
+| 4 | OpenAI pricing / terms change mid-quarter | L | M | Anthropic contract in place; API abstraction layer allows swap in <1 week |
+| 5 | Action items extraction accuracy low → user distrust | M | M | Confidence threshold surfacing; manager always can edit; iterative prompt tuning |
+
+## Rollout Plan
+
+**Phase 1 — Internal Dogfood (Week 6-7):**
+- TeamFlow employees use feature internally
+- Top 5 issues identified + resolved pre-beta
+
+**Phase 2 — Beta (Week 8-10):**
+- 10 Design Partner customers (curated mix: SMB 4, mid-market 4, enterprise 2)
+- Weekly customer interviews + survey
+- NPS target ≥45 during beta
+
+**Phase 3 — Limited GA (Week 11-12):**
+- Feature flag exposed to 25% of AI-tier-upgraded customers
+- Monitor guardrails; gate full rollout on green metrics
+
+**Phase 4 — Full GA (Week 13+):**
+- Feature flag to 100% of AI tier
+- Sales and CS enabled; marketing campaign triggered
+- Tier upgrade funnel open for non-AI-tier customers
+
+**Rollback triggers:**
+- Error rate >2% for >15 minutes → feature flag disable
+- p95 latency >2 min for >30 minutes → disable
+- Support ticket spike 5× baseline → investigation + possible disable
+- Critical privacy incident → immediate disable + customer notification
+
+## Dependencies
+
+- **Platform team:** Event instrumentation schema (Week 4 delivery) — blocker for Week 6 dogfood
+- **Security team:** SOC 2 Type II audit sign-off (Week 10) — blocker for Enterprise deals, not for GA
+- **Legal:** DPA updates sent to customers (Week 8)
+- **Sales:** Enablement training session (Week 11)
+- **CS:** Runbook + training (Week 10)
+- **Marketing:** Landing page + email campaign (Week 11)
+
+## Open Questions
+
+1. Pricing — Team Tier +$8/seat confirmed? (Owner: VP Sales, by Week 3)
+2. Enterprise customers — BYOK encryption delivery path? (Owner: Eng Lead, by Week 5)
+3. Audit log retention — default 365 days, customer override? (Owner: Legal, by Week 4)
+4. Beta customer selection finalized? (Owner: CS Lead, by Week 7)
+
+---
+
+## Appendix A: Full User Stories List (14 stories)
+[Link to Stories spreadsheet]
+
+## Appendix B: Design References
+[Figma link]
+
+## Appendix C: Mediator Synthesis Rationale
+[Link to session-4-handoff.md]
+
+## Appendix D: Metric Plan Details
+[Link to metric-plan.md by Data Analyst]
+```
+
+> **PRD lesson:** every section traces back to Discovery / Strategy / Mediator artifacts — this prevents the "we discussed it somewhere" syndrome. NFR in B2B SaaS is critical — privacy + compliance take up more space than performance. Rollback triggers are numeric — not «if it breaks», but «error rate >2% for >15 min». Non-goals are explicit to prevent scope creep in implementation.

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/claude.json

@@ -0,0 +1,16 @@
+{
+  "name": "prd-template-reference",
+  "display_name": "PRD Template Reference",
+  "description": "Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout — reference (code & worked-example library)",
+  "default_prompt": "Use $prd-template-reference when the task relates to the skill \"PRD Template Reference\".",
+  "triggers": [
+    "prd-template",
+    "PRD",
+    "product requirements document"
+  ],
+  "capabilities": [
+    "prd-template-reference"
+  ],
+  "tools": [],
+  "implicit_invocation": true
+}

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/copilot.json

@@ -0,0 +1,16 @@
+{
+  "name": "prd-template-reference",
+  "display_name": "PRD Template Reference",
+  "description": "Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout — reference (code & worked-example library)",
+  "default_prompt": "Use $prd-template-reference when the task relates to the skill \"PRD Template Reference\".",
+  "triggers": [
+    "prd-template",
+    "PRD",
+    "product requirements document"
+  ],
+  "capabilities": [
+    "prd-template-reference"
+  ],
+  "tools": [],
+  "implicit_invocation": true
+}

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/gemini.json

@@ -0,0 +1,16 @@
+{
+  "name": "prd-template-reference",
+  "display_name": "PRD Template Reference",
+  "description": "Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout — reference (code & worked-example library)",
+  "default_prompt": "Use $prd-template-reference when the task relates to the skill \"PRD Template Reference\".",
+  "triggers": [
+    "prd-template",
+    "PRD",
+    "product requirements document"
+  ],
+  "capabilities": [
+    "prd-template-reference"
+  ],
+  "tools": [],
+  "implicit_invocation": true
+}

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/kimi.yaml

@@ -0,0 +1,15 @@
+name: prd-template-reference
+display_name: PRD Template Reference
+description: Product Requirements Document template for B2B SaaS — problem,
+  solution, stories, AC, NFR, metrics, risks, rollout — reference (code &
+  worked-example library)
+default_prompt: Use $prd-template-reference when the task relates to the skill
+  "PRD Template Reference".
+triggers:
+  - prd-template
+  - PRD
+  - product requirements document
+capabilities:
+  - prd-template-reference
+tools: []
+implicit_invocation: true

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/openai.yaml

@@ -0,0 +1,10 @@
+interface:
+  display_name: PRD Template Reference
+  short_description: Product Requirements Document template for B2B SaaS —
+    problem, solution, stories, AC, NFR, metrics, risks, rollout
+  default_prompt: Use $prd-template-reference when the task relates to the skill
+    "PRD Template Reference".
+dependencies:
+  tools: []
+policy:
+  allow_implicit_invocation: true

package/domains/product/locales/en/.agents/skills/prd-template-reference/agents/qwen.json

@@ -0,0 +1,16 @@
+{
+  "name": "prd-template-reference",
+  "display_name": "PRD Template Reference",
+  "description": "Product Requirements Document template for B2B SaaS — problem, solution, stories, AC, NFR, metrics, risks, rollout — reference (code & worked-example library)",
+  "default_prompt": "Use $prd-template-reference when the task relates to the skill \"PRD Template Reference\".",
+  "triggers": [
+    "prd-template",
+    "PRD",
+    "product requirements document"
+  ],
+  "capabilities": [
+    "prd-template-reference"
+  ],
+  "tools": [],
+  "implicit_invocation": true
+}