arkaos 2.15.0 → 2.16.0

package/departments/ops/skills/operations/references/clickup-ops.md

@@ -0,0 +1,224 @@
+# arka-operations — ClickUp, Tasks, Automation, Channels
+
+Referenced from SKILL.md. Read only when needed.
+
+## /ops onboard-client <name>
+
+**Step 1: Create Client Folder in Obsidian**
+
+Create the directory structure:
+- `WizardingCode/Operations/Clients/<name>/Onboarding.md`
+
+**Step 2: Generate Onboarding Checklist**
+
+**File:** `WizardingCode/Operations/Clients/<name>/Onboarding.md`
+```markdown
+---
+type: onboarding
+department: operations
+title: "<name> — Client Onboarding"
+client: "<name>"
+date_created: <YYYY-MM-DD>
+status: in-progress
+tags:
+  - "client"
+  - "onboarding"
+---
+
+# <name> — Client Onboarding
+
+## Client Info
+- **Company:** <name>
+- **Contact:** [TBD]
+- **Email:** [TBD]
+- **Project:** [TBD]
+- **Start date:** <YYYY-MM-DD>
+
+## Onboarding Checklist
+
+### Admin
+- [ ] Contract signed
+- [ ] Invoice sent (first payment)
+- [ ] Add to ClickUp workspace
+- [ ] Create project in ClickUp
+- [ ] Share Google Drive folder
+
+### Technical
+- [ ] Collect access credentials (hosting, DNS, APIs)
+- [ ] Set up project repository
+- [ ] Scaffold project (`/dev scaffold`)
+- [ ] Configure MCPs for project
+- [ ] Create PROJECT.md
+
+### Communication
+- [ ] Welcome email sent
+- [ ] Kickoff meeting scheduled
+- [ ] Communication channels established
+- [ ] Reporting cadence agreed
+
+### Knowledge
+- [ ] Brand guidelines collected
+- [ ] Existing content/assets received
+- [ ] Competitor list documented
+- [ ] Project brief approved
+
+---
+*Part of the [[WizardingCode MOC]]*
+```
+
+**Step 3: Create ClickUp Tasks**
+- Use ClickUp MCP to create a task list mirroring the onboarding checklist
+- Assign due dates based on a 2-week onboarding timeline
+
+**Step 4: Report**
+```
+═══ ARKA OPS — Client Onboarding Created ═══
+Client:      <name>
+Checklist:   <count> items
+ClickUp:     Tasks created
+Obsidian:    WizardingCode/Operations/Clients/<name>/Onboarding.md
+═════════════════════════════════════════════
+```
+
+## /ops standup
+
+**Step 1: Scan Active Projects**
+- Read `projects/` directory for active project context
+- Check each PROJECT.md for recent updates
+
+**Step 2: Check ClickUp**
+- Use ClickUp MCP to get tasks due today and overdue tasks
+- Get tasks completed yesterday
+
+**Step 3: Check Calendar**
+- Use Google Calendar MCP to get today's meetings and deadlines
+
+**Step 4: Generate Standup Report**
+
+**Output (display, not saved):**
+```
+═══ ARKA OPS — Daily Standup ═══
+Date: <YYYY-MM-DD>
+
+📋 Tasks Due Today
+- [Task 1] — [project]
+- [Task 2] — [project]
+
+⚠️  Overdue
+- [Task] — [days overdue] — [project]
+
+✅ Completed Yesterday
+- [Task 1] — [project]
+- [Task 2] — [project]
+
+📅 Today's Calendar
+- [HH:MM] [Meeting/Event]
+- [HH:MM] [Meeting/Event]
+
+🔥 Blockers
+- [Any identified blockers]
+
+═════════════════════════════════
+```
+
+## /ops automate <process>
+
+**Step 1: Analyze Current Process**
+- Ask user to describe the current manual process step by step
+- Identify frequency, time spent, and pain points
+
+**Step 2: Document Current State**
+- Map all steps, decision points, and handoffs
+- Identify which steps are repetitive and automatable
+
+**Step 3: Design Automation**
+- Propose automation approach (ClickUp automations, scripts, email templates, etc.)
+- Identify tools and integrations needed
+- Estimate time saved per occurrence
+
+**Step 4: Create SOP in Obsidian**
+
+**File:** `WizardingCode/Operations/Automations/<process>.md`
+```markdown
+---
+type: automation
+department: operations
+title: "<process> — Automation SOP"
+date_created: <YYYY-MM-DD>
+status: draft
+time_saved: "<estimate per occurrence>"
+tags:
+  - "automation"
+  - "sop"
+  - "<process-kebab-case>"
+---
+
+# <process> — Automation SOP
+
+## Current Process (Before)
+1. [Manual step]
+2. [Manual step]
+
+## Automated Process (After)
+1. [Automated/simplified step]
+2. [Automated/simplified step]
+
+## Implementation
+- **Tools:** [ClickUp, Gmail, scripts, etc.]
+- **Setup steps:** [numbered list]
+- **Triggers:** [what kicks off the automation]
+
+## Estimated Impact
+- Time saved per occurrence: [X min]
+- Frequency: [X times/week]
+- Total weekly savings: [X hours]
+
+---
+*Part of the [[WizardingCode MOC]]*
+```
+
+**Step 5: Report**
+```
+═══ ARKA OPS — Automation Designed ═══
+Process:     <process>
+Steps:       <before count> → <after count>
+Time saved:  <estimate per occurrence>
+Obsidian:    WizardingCode/Operations/Automations/<process>.md
+═══════════════════════════════════════
+```
+
+## /ops channel add <platform> <channel-id>
+
+**Supported platforms:** `slack`, `discord`, `whatsapp`, `teams`
+
+1. Read `knowledge/channels-config.json`
+2. Add entry: `"<platform>": { "channel_id": "<channel-id>", "added": "<date>" }`
+3. If no default notification channel exists, set this as the default
+4. Write updated config back
+5. Confirm: "Channel added. Use `/ops notify` to send messages."
+
+## /ops channel list
+
+1. Read `knowledge/channels-config.json`
+2. Display all configured channels with their platform and channel ID
+3. Mark the default notification channel with `[DEFAULT]`
+
+## /ops channel remove <platform>
+
+1. Read `knowledge/channels-config.json`
+2. Remove the specified platform entry
+3. If the removed channel was the default, clear the default
+4. Write updated config back
+
+## /ops notify <message>
+
+1. Read `knowledge/channels-config.json` → get default notification channel
+2. If no default set, warn and list available channels
+3. Use the appropriate MCP (Slack, Discord, WhatsApp, or Teams) to send the message
+4. Confirm delivery
+
+## /ops broadcast <message>
+
+1. Read `knowledge/channels-config.json` → get all configured channels
+2. For each channel, use the appropriate MCP to send the message
+3. Report delivery status per channel

package/departments/ops/skills/update/SKILL.md

@@ -13,57 +13,31 @@ Hybrid sync engine: Python handles deterministic operations (MCPs, settings, des
 /arka update
 ```
 
-## How It Works
+## Commands
 
-### Phase 1-3 + 5: Deterministic Sync (Python)
+| Command | Description |
+|---------|-------------|
+| `/arka update` | Full sync — run engine, dispatch AI subagent for skills, write state, report |
 
-Run the sync engine:
-```bash
-cd $ARKAOS_ROOT && python -m core.sync.engine --home ~/.arkaos --skills ~/.claude/skills --output json
-```
-
-The engine:
-1. **Phase 1 (Manifest):** Reads sync-state.json, compares versions, loads feature registry
-2. **Phase 2 (Discovery):** Finds all projects from 3 sources (descriptors, filesystem, ecosystems)
-3. **Phase 3a (MCP Sync):** Updates .mcp.json per project based on registry + stack
-4. **Phase 3b (Settings Sync):** Aligns settings.local.json with .mcp.json
-5. **Phase 3c (Descriptors):** Auto-pause inactive, archive missing, update stacks
-6. **Phase 5 (State):** Writes sync-state.json and returns JSON report
-
-### Phase 4: Intelligent Sync (AI Subagent)
+## Orchestration (Summary)
 
-After the Python engine completes, dispatch ONE subagent to update ecosystem skills:
+1. **Phases 1–3 + 5 (deterministic, Python):** Run the engine:
+   ```bash
+   cd $ARKAOS_ROOT && python -m core.sync.engine --home ~/.arkaos --skills ~/.claude/skills --output json
+   ```
+   Handles manifest, discovery, MCP sync, settings sync, descriptors, and writes `~/.arkaos/sync-state.json`.
 
-**Subagent input:**
-- The JSON report from the engine (which ecosystems exist)
-- The feature registry files from `core/sync/features/*.yaml` (or `~/.arkaos/config/sync/features/*.yaml`)
+2. **Phase 4 (intelligent, AI subagent):** After the engine completes, dispatch ONE subagent with the engine's JSON report + the feature registry (`core/sync/features/*.yaml`). The subagent injects/removes feature sections in each `~/.claude/skills/arka-{ecosystem}/SKILL.md` while preserving all custom content.
 
-**Subagent task:**
-For each ecosystem skill (`~/.claude/skills/arka-{ecosystem}/SKILL.md`):
-1. Read the SKILL.md
-2. For each feature YAML where `deprecated_in` is null:
-   - Search SKILL.md for `detection_pattern`
-   - If NOT found → inject `content` after the last existing feature section, or after the "Commands" table if no feature sections exist yet (before "Orchestration Workflows" section)
-3. For each feature where `deprecated_in` is set:
-   - Search for and remove the section
-4. PRESERVE all custom content: commands, architecture, tech stack, business descriptions
+3. **Report:** Display the formatted summary returned by the engine.
 
-### Report
+## Error Handling (Summary)
 
-Display the formatted report from the engine output. The engine's text output format:
-```
-═══════════════════════════════════════════════════════
-  ArkaOS Sync Complete — v2.14.0 → v2.15.0
-═══════════════════════════════════════════════════════
-  MCPs:         22 synced (8 updated, 14 unchanged)
-  Settings:     22 synced (8 updated, 14 unchanged)
-  Descriptors:  5 synced (1 updated, 4 unchanged)
-  Skills:       3 ecosystems synced (2 updated, 1 unchanged)
-  ...
-```
+- Python engine fails → report error, do not proceed to AI phase.
+- AI subagent fails → deterministic sync already completed, report partial success.
+- Individual project errors never stop other projects from syncing.
 
-## Error Handling
+## References
 
-- If Python engine fails: report the error, do not proceed to AI phase
-- If AI subagent fails: deterministic sync already completed, report partial success
-- Individual project errors don't stop other projects from syncing
+- `references/sync-engine.md` — Python engine phases (manifest, discovery, MCP/settings/descriptor sync, state), feature registry format, key paths
+- `references/workflows.md` — 2-step update flow, Phase 4 AI subagent instructions, report format, full error-handling table

package/departments/ops/skills/update/references/sync-engine.md

@@ -0,0 +1,93 @@
+# arka-update — sync engine
+
+Referenced from SKILL.md. Read only when needed.
+
+## Python Engine Invocation
+
+Run the sync engine from the ArkaOS repo root:
+
+```bash
+cd $ARKAOS_ROOT && python -m core.sync.engine --home ~/.arkaos --skills ~/.claude/skills --output json
+```
+
+The engine handles deterministic operations and returns a JSON report.
+
+## Engine Phases
+
+### Phase 1 — Manifest
+- Reads `~/.arkaos/sync-state.json` (last synced version)
+- Compares against current `VERSION` from the repo
+- Loads feature registry from `core/sync/features/*.yaml` (or `~/.arkaos/config/sync/features/*.yaml`)
+- Builds change manifest (agents, MCPs, workflows, features added/removed/modified)
+
+### Phase 2 — Discovery
+Combines three sources to find all ArkaOS-managed projects:
+1. **Ecosystem registry:** `~/.claude/skills/arka/knowledge/ecosystems.json`
+2. **Filesystem scan:** projectsDir from `~/.arkaos/profile.json` (e.g. `~/Herd/`, `~/Work/`, `~/AIProjects/`). A subdir counts as an ArkaOS project if it has any of: `.mcp.json`, `.claude/`, `CLAUDE.md`.
+3. **Descriptors:** `~/.claude/skills/arka/projects/*.md` and `~/.claude/skills/arka/projects/*/PROJECT.md`
+
+Deduplicate by absolute path. Skip projects whose path does not exist.
+
+### Phase 3a — MCP Sync
+For each project: regenerate `.mcp.json` based on the registry + detected stack.
+
+Base MCPs (all projects): `arka-prompts`, `context7`, `obsidian`, `clickup`, `memory-bank`, `playwright`, `gh-grep`.
+
+Stack additions:
+- Laravel: `laravel-boost`, `serena`, `sentry`
+- Nuxt/Vue: nuxt-specific MCPs from registry
+- Shopify: `shopify-dev`
+- PostgreSQL / Supabase: matching MCP if in registry
+
+Rules:
+- `serena` MCP: update `--project` arg to the project's absolute path
+- All MCPs: copy exact command/args/env from registry; resolve `~`/`${HOME}` to real path
+- **Preserve** project-specific MCPs not in the registry (user-added)
+
+### Phase 3b — Settings Sync
+Align `.claude/settings.local.json` with the updated `.mcp.json`:
+- `enabledMcpjsonServers` matches the final MCP server list
+- `enableAllProjectMcpServers: true`
+- **Preserve** any custom `permissions` the user has configured
+- If the file does not exist, create it with permissions `["Read","Grep","Glob","WebFetch"]` plus the MCP list
+
+### Phase 3c — Descriptors
+For each descriptor:
+- If `path:` no longer exists: `status: archived` + note "Path not found on filesystem"
+- Detect stack from `composer.json` / `package.json` / `pyproject.toml` and update `stack:`
+- Git activity: last commit > 30 days ago + active → paused; < 7 days + paused → active
+- Validate `ecosystem:` matches an existing `~/.claude/skills/arka-*/`
+
+### Phase 5 — State
+Writes `~/.arkaos/sync-state.json`:
+```json
+{
+  "version": "<current VERSION>",
+  "last_sync": "<ISO 8601>",
+  "projects_synced": <count>,
+  "skills_synced": <count>,
+  "errors": [<messages>]
+}
+```
+Returns JSON report for downstream consumption.
+
+## Feature Registry
+
+YAML files under `core/sync/features/*.yaml` (or `~/.arkaos/config/sync/features/*.yaml`). Each feature has:
+- `detection_pattern` — regex/string searched in ecosystem SKILL.md to decide if the feature is already present
+- `content` — the section to inject if missing
+- `deprecated_in` — if set, the matching section is removed
+
+## Key Paths
+
+| Path | Purpose |
+|------|---------|
+| `~/.arkaos/sync-state.json` | Sync state (version, timestamp, counts, errors) |
+| `~/.arkaos/.repo-path` | Points to the ArkaOS npm package directory |
+| `~/.arkaos/profile.json` | User profile (language, market, projectsDir, vaultPath) |
+| `~/.arkaos/install-manifest.json` | Installation metadata |
+| `~/.claude/skills/arka/knowledge/ecosystems.json` | Ecosystem registry |
+| `~/.claude/skills/arka/mcps/registry.json` | MCP server definitions (source of truth) |
+| `~/.claude/skills/arka/mcps/stacks/*.json` | Stack-specific package/MCP configs |
+| `~/.claude/skills/arka/projects/` | Project descriptor files |
+| `~/.claude/skills/arka-*/SKILL.md` | Ecosystem skill definitions |

package/departments/ops/skills/update/references/workflows.md

@@ -0,0 +1,84 @@
+# arka-update — workflows
+
+Referenced from SKILL.md. Read only when needed.
+
+## The 2-Step Update Flow
+
+ArkaOS updates happen in two steps.
+
+### Step 1: Core Update (terminal, outside Claude Code)
+
+```bash
+npx arkaos@latest update
+```
+
+Downloads the latest ArkaOS npm package and:
+- Updates Python dependencies
+- Copies new hook scripts (SessionStart, UserPromptSubmit, PostToolUse, PreCompact, CwdChanged)
+- Updates the `/arka` skill and `arka-claude` CLI wrapper
+- Resets sync state → triggers `[arka:update-available]` warning on next session
+
+### Step 2: Project Sync (inside Claude Code)
+
+```
+/arka update
+```
+
+AI-powered sync that updates ecosystem skills, project descriptors, MCP configs, and Claude settings.
+
+### When To Run
+
+- After `npx arkaos@latest update` bumps the core version
+- When the SessionStart hook shows `[arka:update-available]`
+- Manually, any time, to force a full sync
+
+## Hybrid Orchestration
+
+Phases 1–3 + 5 run via the Python engine (see `sync-engine.md`). Phase 4 runs as ONE AI subagent to handle intelligent ecosystem-skill text updates.
+
+### Phase 4 — Intelligent Sync (AI Subagent)
+
+After the Python engine completes, dispatch ONE subagent.
+
+**Subagent input:**
+- The JSON report from the engine (list of ecosystems that exist)
+- The feature registry files from `core/sync/features/*.yaml` (or `~/.arkaos/config/sync/features/*.yaml`)
+
+**Subagent task — for each `~/.claude/skills/arka-{ecosystem}/SKILL.md`:**
+1. Read the SKILL.md.
+2. For each feature YAML where `deprecated_in` is null:
+   - Search SKILL.md for `detection_pattern`.
+   - If NOT found: inject `content` after the last existing feature section, or after the "Commands" table if no feature sections exist (before "Orchestration Workflows").
+3. For each feature where `deprecated_in` is set:
+   - Locate and remove the matching section.
+4. PRESERVE all custom content: commands, architecture, tech stack, business descriptions, ecosystem-specific workflow details.
+
+### Report
+
+Display the formatted report from the engine output:
+
+```
+═══════════════════════════════════════════════════════
+  ArkaOS Sync Complete — v2.14.0 → v2.15.0
+═══════════════════════════════════════════════════════
+  MCPs:         22 synced (8 updated, 14 unchanged)
+  Settings:     22 synced (8 updated, 14 unchanged)
+  Descriptors:  5 synced (1 updated, 4 unchanged)
+  Skills:       3 ecosystems synced (2 updated, 1 unchanged)
+  ...
+```
+
+## Error Handling
+
+| Scenario | Action |
+|----------|--------|
+| Python engine fails | Report error; do NOT proceed to AI phase |
+| AI subagent fails | Deterministic sync already completed; report partial success |
+| Individual project error | Other projects continue; the failure is recorded in `sync-state.json` errors |
+| Project path not found | Skip project, warn, do not delete its descriptor |
+| No stack detectable | Use generic MCPs only (`arka-prompts`, `context7`, `clickup`, `obsidian`) |
+| Ecosystem skill has manual customizations | Update only structural sections; preserve all custom content |
+| First sync (no `sync-state.json` or version is `pending-sync` / `none`) | Full sync without diff, create `sync-state.json` |
+| Version downgrade (sync-state version > current VERSION) | Warn in report, sync anyway |
+| `.mcp.json` has MCPs not in registry | Preserve them (user-added, project-specific) |
+| `.claude/settings.local.json` has custom permissions | Preserve them (user-configured) |

package/departments/ops/workflows/lean-audit.yaml

@@ -43,6 +43,7 @@ phases:
 
   - id: quality-gate
     name: Quality Gate
+    model_override: opus
     description: Validate analysis quality
     agents:
       - agent_id: cqo-marta

package/departments/org/SKILL.md

@@ -32,3 +32,16 @@ allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent, WebFetch, WebSearch]
 | **Sofia** (COO) | 0 | S+C | Org strategy, cross-department coordination, scaling |
 
 ## Frameworks: Team Topologies (Skelton/Pais), Spotify Model (Kniberg), Five Dysfunctions (Lencioni), OKRs (Doerr), Netflix Culture, Conway's Law, RACI Matrix
+
+## Model Selection
+
+When dispatching subagent work via the Task tool, include the `model` parameter from the target agent's YAML `model:` field:
+
+- Agent YAMLs at `departments/*/agents/*.yaml` have `model: opus | sonnet | haiku`
+- Quality Gate dispatch (Marta/Eduardo/Francisca) ALWAYS uses `model: opus` — NON-NEGOTIABLE
+- Default to `sonnet` if the agent YAML has no `model` field
+- Mechanical tasks (commit messages, routing, keyword extraction) use `model: haiku`
+
+Example Task tool call:
+
+    Task(description="...", subagent_type="general-purpose", model="sonnet", prompt="...")

package/departments/org/agents/coo.yaml

@@ -3,6 +3,7 @@ name: Sofia
 role: Chief Operations Officer
 department: org
 tier: 0
+model: opus
 
 behavioral_dna:
   disc:

package/departments/org/agents/org-designer.yaml

@@ -3,6 +3,7 @@ name: Pedro M.
 role: Organizational Designer
 department: org
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/org/agents/people-ops.yaml

@@ -3,6 +3,7 @@ name: Carla
 role: People Operations Specialist
 department: org
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/org/workflows/design.yaml

@@ -25,6 +25,7 @@ phases:
     type: auto
 - id: design
   name: New Structure
+  model_override: opus
   agents:
   - agent_id: org-designer-pedro-m
     role: Propose new topology with team types and interaction modes
@@ -33,6 +34,7 @@ phases:
     description: User approves proposed structure
 - id: quality-gate
   name: Quality Gate
+  model_override: opus
   agents:
   - agent_id: cqo-marta
     role: Orchestrate

package/departments/pm/SKILL.md

@@ -37,3 +37,16 @@ allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent, WebFetch, WebSearch]
 | **Jorge** (SM) | 2 | S+I | Facilitation, flow metrics, retros, coaching |
 
 ## Frameworks: Scrum (Sutherland), Kanban (Anderson), Shape Up (Singer), Continuous Discovery (Torres), OST, RICE/WSJF, Monte Carlo (Vacanti), User Story Mapping (Patton), Impact Mapping (Adzic), INVEST
+
+## Model Selection
+
+When dispatching subagent work via the Task tool, include the `model` parameter from the target agent's YAML `model:` field:
+
+- Agent YAMLs at `departments/*/agents/*.yaml` have `model: opus | sonnet | haiku`
+- Quality Gate dispatch (Marta/Eduardo/Francisca) ALWAYS uses `model: opus` — NON-NEGOTIABLE
+- Default to `sonnet` if the agent YAML has no `model` field
+- Mechanical tasks (commit messages, routing, keyword extraction) use `model: haiku`
+
+Example Task tool call:
+
+    Task(description="...", subagent_type="general-purpose", model="sonnet", prompt="...")

package/departments/pm/agents/pm-director.yaml

@@ -3,6 +3,7 @@ name: Carolina
 role: Product Manager
 department: pm
 tier: 1
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/pm/agents/product-owner.yaml

@@ -3,6 +3,7 @@ name: Sara
 role: Product Owner
 department: pm
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/pm/agents/scrum-master.yaml

@@ -3,6 +3,7 @@ name: Jorge
 role: Scrum Master / Agile Coach
 department: pm
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/quality/SKILL.md

@@ -64,3 +64,16 @@ Any Department Workflow:
 | **REJECTED** | One or more issues found | Loop back with specific issue list |
 
 There is no "APPROVED WITH CAVEATS". It's binary. Fix issues first.
+
+## Model Selection
+
+When dispatching subagent work via the Task tool, include the `model` parameter from the target agent's YAML `model:` field:
+
+- Agent YAMLs at `departments/*/agents/*.yaml` have `model: opus | sonnet | haiku`
+- Quality Gate dispatch (Marta/Eduardo/Francisca) ALWAYS uses `model: opus` — NON-NEGOTIABLE
+- Default to `sonnet` if the agent YAML has no `model` field
+- Mechanical tasks (commit messages, routing, keyword extraction) use `model: haiku`
+
+Example Task tool call:
+
+    Task(description="...", subagent_type="general-purpose", model="sonnet", prompt="...")

package/departments/quality/agents/copy-director.yaml

@@ -3,6 +3,7 @@ name: Eduardo
 role: Copy & Language Director
 department: quality
 tier: 0
+model: opus
 
 behavioral_dna:
   disc:

package/departments/quality/agents/cqo.yaml

@@ -3,6 +3,7 @@ name: Marta
 role: Chief Quality Officer
 department: quality
 tier: 0
+model: opus
 
 behavioral_dna:
   disc:

package/departments/quality/agents/tech-director.yaml

@@ -3,6 +3,7 @@ name: Francisca
 role: Technical & UX Quality Director
 department: quality
 tier: 0
+model: opus
 
 behavioral_dna:
   disc:

package/departments/saas/SKILL.md

@@ -62,3 +62,16 @@ allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent, WebFetch, WebSearch]
 | Value-Based Pricing | Patrick Campbell | Pricing based on willingness to pay |
 | Customer Success Lifecycle | Lincoln Murphy | Onboard, adopt, expand, renew, advocate |
 | AARRR / RARRA | Dave McClure | Growth funnel metrics |
+
+## Model Selection
+
+When dispatching subagent work via the Task tool, include the `model` parameter from the target agent's YAML `model:` field:
+
+- Agent YAMLs at `departments/*/agents/*.yaml` have `model: opus | sonnet | haiku`
+- Quality Gate dispatch (Marta/Eduardo/Francisca) ALWAYS uses `model: opus` — NON-NEGOTIABLE
+- Default to `sonnet` if the agent YAML has no `model` field
+- Mechanical tasks (commit messages, routing, keyword extraction) use `model: haiku`
+
+Example Task tool call:
+
+    Task(description="...", subagent_type="general-purpose", model="sonnet", prompt="...")

package/departments/saas/agents/cs-manager.yaml

@@ -3,6 +3,7 @@ name: Patricia
 role: Customer Success Manager
 department: saas
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc:

package/departments/saas/agents/growth-engineer.yaml

@@ -3,6 +3,7 @@ name: Andre S.
 role: Growth Engineer
 department: saas
 tier: 2
+model: sonnet
 
 behavioral_dna:
   disc: