@monoes/monomindcli 2.16.2 → 2.16.4

package/.claude/skills/mastermind-review/SKILL.md

@@ -39,7 +39,7 @@ Assess the prompt to determine execution mode:
 **Simple (direct execution):** Single file or single artifact:
 - "Review this function for bugs"
 - "Check this paragraph for clarity"
-→ Use a single Code Reviewer or content reviewer agent. Skip manager delegation.
+→ Use a single `reviewer` agent (for code or content). Skip manager delegation.
 
 **Complex (spawn Review Manager agent):** Any of these:
 - Full codebase or module audit
@@ -116,19 +116,24 @@ SCOPE: [exact files, URLs, documents, or system surfaces in scope]
 CONSTRAINTS: [known acceptable risks, existing decisions not to revisit, standards to apply]
 SUCCESS CRITERIA:
 - [ ] [checkable item — e.g. \"all critical issues documented\"]
-AGENT: [Code Reviewer | Security Engineer | analyst | Accessibility Auditor | UX Researcher]
+AGENT: [specialist picked in STEP 3]
 SWARM: mesh 4 gossip
 DEPENDENCIES: [task IDs or \"none\"]
 OUTPUT FORMAT: unified output schema"
 ```
 
 STEP 3 — EXECUTE
-Spawn one Task agent per review angle (mesh topology — reviewers share findings):
-- Code quality: subagent_type "Code Reviewer"
-- Security: subagent_type "Security Engineer"
-- Architecture: subagent_type "Software Architect"
-- Analytics/metrics: subagent_type "Analytics Reporter"
-- Accessibility: subagent_type "Accessibility Auditor"
+Spawn one Task agent per review angle (mesh topology — reviewers share findings).
+Pick one specialist per review angle from the shared agent index (the Jev decision model when configured, keyword ranking otherwise); it only returns agents that exist. In order: the prompt's `[PICK]` line when it fits this review angle; else `mcp__monomind__pick({ task: "<review angle>: <scope>", kind: "agents", top: 1 })` → `agents.ranked[0].name`; else the local CLI through the version-checked `mmpick` helper from `mastermind-agent-select/SKILL.md` (never npx):
+```bash
+mmpick -t "<review angle>: <scope>" --top 1 | jq -r '.agents.ranked[0].name // empty'
+```
+Use the returned name as that review angle's subagent_type. If nothing is returned, use the default below.
+- Code quality: "reviewer"
+- Security: "Security Engineer"
+- Architecture: "Software Architect"
+- Accessibility: "Accessibility Auditor"
+- Any other angle: "reviewer"
 
 **If `monofence_check = true`** — add a monofence-ai self-validation step to the Security Engineer's briefing:
 ```
@@ -227,7 +232,7 @@ For simple tasks (single reviewer, single artifact):
 | Task Type | Agent | Swarm |
 |---|---|---|
 | Full multi-angle review | reviewer + specialists | mesh 4 gossip balanced |
-| Security audit | Security Engineer + Code Reviewer | hive-mind hierarchical-mesh byzantine 6 |
-| Code review only | Code Reviewer | hierarchical 3 raft specialized |
+| Security audit | Security Engineer + reviewer | hive-mind hierarchical-mesh byzantine 6 |
+| Code review only | reviewer | hierarchical 3 raft specialized |
 | Strategy review | analyst + researcher | mesh 3 gossip balanced |
-| Content review | Code Reviewer (content) | single agent |
+| Content review | reviewer (content) | single agent |

package/.claude/skills/mastermind-routine-detail/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-routine-detail
 description: Mastermind routine-detail — deep inspection and management of a single routine: trigger config (schedule/webhook), variables, concurrency/catchup policies, run history, webhook rotation, and revision tracking.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
 # Mastermind Routine Detail
 
-This skill is invoked by `mastermind:routine-detail` or directly via `/mastermind:routine-detail`.
+This skill is invoked by `mastermind:routine-detail` or directly via `/mastermind-routine-detail`.
 
 ---
 

package/.claude/skills/mastermind-routines/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-routines
 description: Mastermind routines — schedule recurring tasks that trigger agent heartbeats on a cron-like schedule within a running org. Configure concurrency policy and catch-up behavior.
 type: domain-skill
 default_mode: confirm
+pick: low
 ---
 
 # Mastermind Routines
 
-This skill is invoked by `mastermind:routines` or directly via `/mastermind:routines`.
+This skill is invoked by `mastermind:routines` or directly via `/mastermind-routines`.
 
 ---
 
@@ -103,7 +104,7 @@ echo "Routine added: $routine_id"
 echo "Schedule: $schedule"
 echo ""
 echo "NOTE: To activate this routine, use ScheduleWakeup with:"
-echo "  prompt: '/mastermind:routines --action trigger --org $org_name --routine-id $routine_id'"
+echo "  prompt: '/mastermind-routines --action trigger --org $org_name --routine-id $routine_id'"
 echo "  delaySeconds: <seconds until next scheduled run>"
 ```
 

package/.claude/skills/mastermind-runorg/SKILL.md

@@ -1,13 +1,13 @@
 ---
 name: mastermind-runorg
-description: Start a saved org via the Org Runtime v2 daemon (monomind org run/serve). Migrates v1-shaped configs first. The legacy prompt-orchestrated path lives in runorgv1.
+description: Start a saved org via the Org Runtime daemon (monomind org run/serve). Converts legacy-format config files with monomind org migrate first.
 type: domain-skill
 default_mode: auto
 ---
 
-# Mastermind Runorg (v2)
+# Mastermind Runorg
 
-Starts orgs exclusively through the v2 SDK daemon. Every role becomes a live
+Starts orgs through the Org Runtime SDK daemon. Every role becomes a live
 SDK session; events reach the dashboard through the daemon's own forwarder —
 no curl emissions, no delivery gaps.
 
@@ -15,17 +15,18 @@ no curl emissions, no delivery gaps.
 
 1. **Resolve the org.** `org_name` from params. List available orgs when missing:
    `monomind org list`.
-2. **Shape-detect v1.** Do NOT rely on `monomind org validate` to catch v1
-   configs — a v1-shaped config with no structural violations (unique role
-   ids, one root, resolvable `reports_to`, parseable schedule) *passes*
-   schema validation even though it's still v1. Detect v1-ness directly:
+2. **Detect the legacy config format.** Do NOT rely on `monomind org validate`
+   to catch legacy-format config files — one with no structural violations
+   (unique role ids, one root, resolvable `reports_to`, parseable schedule)
+   *passes* schema validation even though it still carries legacy fields.
+   Detect them directly:
    `jq 'has("topology") or has("board_id") or has("loop") or ([.roles[]? | has("agent_type")] | any)' .monomind/orgs/<name>.json`
    - `true` → step 3 (migrate first).
    - `false` → step 4 (validate as-is).
-3. **Migrate (v1 configs).** `monomind org migrate <name>` — original is kept
-   as `<name>.v1.json`. In confirm mode ask first; in auto mode migrate and
-   state it. If migration fails, stop and surface the error — do NOT fall back
-   to runorgv1 silently.
+3. **Migrate (legacy-format configs).** `monomind org migrate <name>` rewrites
+   the file in the current format — the original is kept as `<name>.v1.json`.
+   In confirm mode ask first; in auto mode migrate and state it. If migration
+   fails, stop and surface the error.
 4. **Validate.** Run `monomind org validate <name>` on the config (post-migration
    if step 3 ran).
    - Valid → step 5.
@@ -55,8 +56,7 @@ no curl emissions, no delivery gaps.
    `running`. Surface the dashboard link (`<CTRL_URL>/orgs`) and
    `monomind org logs <name> --follow` as the tail command.
 8. **Never** spawn a boss Task agent, create monotask boards, or emit
-   dashboard events manually. If the user explicitly asks for the legacy
-   behavior, direct them to `/mastermind:runorgv1` (deprecated).
+   dashboard events manually. The org daemon is the only way an org runs.
 
 The org runtime does not read `<org>-issues.json`. Work is driven by the org
 definition's roles and goal plus an optional `--task` string.

package/.claude/skills/mastermind-search/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-search
 description: Mastermind search — fuzzy cross-org search across tasks, goals, agents, approvals, routines, projects, and activity log. Returns ranked results with org context.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
 # Mastermind Search
 
-This skill is invoked by `mastermind:search` or directly via `/mastermind:search`.
+This skill is invoked by `mastermind:search` or directly via `/mastermind-search`.
 
 ---
 

package/.claude/skills/mastermind-secrets/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-secrets
 description: Mastermind secrets — manage org-scoped API keys and secrets consumed by agents. Store, rotate, list, and audit secrets without exposing values in logs or state files.
 type: domain-skill
 default_mode: confirm
+pick: low
 ---
 
 # Mastermind Secrets
 
-This skill is invoked by `mastermind:secrets` or directly via `/mastermind:secrets`.
+This skill is invoked by `mastermind:secrets` or directly via `/mastermind-secrets`.
 
 ---
 
@@ -91,7 +92,7 @@ chmod 700 "$secretDir"
 
 # Write value from env var reference — NEVER inline
 if [ -z "${secret_value}" ]; then
-  echo "ERROR: Pass the secret value as an env var: secret_value=\$MY_VAR /mastermind:secrets --action set"
+  echo "ERROR: Pass the secret value as an env var: secret_value=\$MY_VAR /mastermind-secrets --action set"
   exit 1
 fi
 

package/.claude/skills/mastermind-skills/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-skills
 description: Mastermind skills — list, sync, and map skills available to org agents. Scans .claude/skills/ directory and shows which roles have access to which skill domains.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
 # Mastermind Skills
 
-This skill is invoked by `mastermind:skills` or directly via `/mastermind:skills`.
+This skill is invoked by `mastermind:skills` or directly via `/mastermind-skills`.
 
 ---
 

package/.claude/skills/mastermind-stoporg/SKILL.md

@@ -1,8 +1,9 @@
 ---
 name: mastermind-stoporg
-description: Mastermind stoporg — stop a running scheduled org loop by setting its status to "stopped". The next scheduled wakeup will read the status, skip all work, and not reschedule. Loop dies within one interval — no orphaned wakeups.
+description: Mastermind stoporg — stop a running org. Wraps monomind org stop, which writes the stop file the org daemon polls; the daemon exits within about 2 seconds.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
 # Mastermind Stop Org
@@ -36,15 +37,8 @@ orgFile=".monomind/orgs/${org_name}.json"
 }
 ```
 
-Read current status:
-```bash
-current_status=$(jq -r '.status // "no-schedule"' "$orgFile")
-has_schedule=$(jq -r 'if .loop.poll_interval_minutes then "yes" else "no" end' "$orgFile")
-```
-
-If `has_schedule == "no"` — this is a **v2 org** (Org Runtime v2 has `schedule`,
-never `.loop`). Its daemon polls `.monomind/orgs/<name>/stop` every 2s, which is
-exactly what `monomind org stop` writes — use the CLI, not the v1 `.stops/` path:
+The org daemon polls `.monomind/orgs/<name>/stop` every 2s, which is exactly
+what `monomind org stop` writes — use the CLI:
 
 ```bash
 REPO_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
@@ -53,67 +47,19 @@ npx -y monomind@latest org stop "${org_name}" \
   || { mkdir -p ".monomind/orgs/${org_name}"; date -u +%Y-%m-%dT%H:%M:%SZ > ".monomind/orgs/${org_name}/stop"; }
 # Also POST to the control server in case a dashboard-started instance is running
 curl -s -X POST -H "x-monomind-token: $(cat "${REPO_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}/.monomind/dashboard-token" 2>/dev/null || true)" "${CTRL_URL}/api/orgs/${org_name}/stop" >/dev/null 2>&1 || true
-echo "Stop requested for org '${org_name}' (v2 daemon exits within 2s)."
-```
-- Exit.
-
-<!-- LEGACY-ORG-V1: remove this branch when v1 orgs are gone -->
-Everything below this point only runs for v1 orgs — v2 orgs already exited above.
-
-If `current_status == "stopped"`:
-- Print: "Org '<org_name>' is already stopped."
-- Exit.
-
----
-
-<!-- LEGACY-ORG-V1: remove this step when v1 orgs are gone -->
-## Step 2 — Set Status to Stopped and Clear next_run (v1 only)
-
-```bash
-orgFile=".monomind/orgs/${org_name}.json"
-tmp="${orgFile}.tmp"
-# Clear next_run to avoid showing a stale future timestamp in orgstatus
-jq '.status = "stopped" | if .loop then .loop.next_run = null else . end' "$orgFile" > "$tmp" && mv "$tmp" "$orgFile"
-```
-
----
-
-<!-- LEGACY-ORG-V1: remove this step when v1 orgs are gone -->
-## Step 3 — Write Stop File (for any running boss agents, v1 only)
-
-Any persistent boss agent checks for this file at the start of each loop iteration:
-
-```bash
-mkdir -p .monomind/orgs/.stops
-touch ".monomind/orgs/.stops/${org_name}.stop"
-```
-
----
-
-<!-- LEGACY-ORG-V1: remove this step when v1 orgs are gone -->
-## Step 5 — Report to User (v1 only)
-
-```
-✓ Org "<org_name>" stopped.
-
-  Current status: stopped
-  The loop will exit at its next scheduled wakeup without rescheduling.
-  Loop fully dead within: ≤$(jq -r '.loop.poll_interval_minutes // "?"' "$orgFile") minutes
-
-  To restart: /mastermind:runorg --org <org_name>
-  Status:     /mastermind:orgstatus --org <org_name>
+echo "Stop requested for org '${org_name}' (the daemon exits within 2s)."
 ```
 
 ---
 
-## Step 6 — Return Output
+## Step 2 — Return Output
 
 ```yaml
 domain: ops
 status: complete
 decisions:
   - what: "Org <org_name> stopped"
-    why: "status set to 'stopped'; next wakeup exits without rescheduling"
+    why: "stop file written; the org daemon exits within 2s"
     confidence: 1.0
     outcome: shipped
 next_actions:
@@ -123,6 +69,6 @@ next_actions:
 
 ---
 
-## Step 7 — Brain Write (standalone only)
+## Step 3 — Brain Write (standalone only)
 
 If `caller` is not "command", follow mastermind-protocol/SKILL.md Brain Write Procedure for domain `ops`.

package/.claude/skills/mastermind-tasks/SKILL.md

@@ -3,16 +3,17 @@ name: mastermind-tasks
 description: Mastermind tasks — view, create, assign, and move tasks on an org's task board. Supports parent-child chains, goal linkage, and status filtering.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
-> **LEGACY-ORG-V1 — writes to a monotask board, not to `<org>-issues.json`.** This skill's
-> `parent_id`, `status` (`todo | doing | done`), and card model belong to the pre-v2
-> board-backed runner. For v2 issue tracking — including sub-issues — use
-> `/mastermind:issues --action create --parent-id <id>`.
+> **Writes to a monotask board, not to `<org>-issues.json`.** This skill only works
+> on an org config that still carries a monotask `board_id` (the legacy config
+> format); the Org Runtime never reads that board. For org issue tracking —
+> including sub-issues — use `/mastermind-issues --action create --parent-id <id>`.
 
 # Mastermind Tasks
 
-This skill is invoked by `mastermind:tasks` or directly via `/mastermind:tasks`.
+This skill is invoked by `mastermind:tasks` or directly via `/mastermind-tasks`.
 
 ---
 
@@ -40,16 +41,14 @@ If `caller` is not "command", load brain context following mastermind-protocol/S
 ## Step 1 — Load Org Config
 
 ```bash
-# LEGACY-ORG-V1: this whole board/column lookup is the pre-v2 board-backed
-# task model — boards belong to the legacy v1 runner, see runorgv1.
+# board/column ids only exist on legacy-format org configs
 orgFile=".monomind/orgs/${org_name}.json"
 board_id=$(jq -r '.board_id // empty' "$orgFile")
 todo_col=$(jq -r '.todo_col_id // empty' "$orgFile")
 doing_col=$(jq -r '.doing_col_id // empty' "$orgFile")
 done_col=$(jq -r '.done_col_id // empty' "$orgFile")
 
-[ -z "$board_id" ] && { echo "ERROR: org config missing board_id — boards belong to the legacy v1 runner, see runorgv1."; exit 1; }
-# end LEGACY-ORG-V1
+[ -z "$board_id" ] && { echo "ERROR: org config has no board_id — the Org Runtime does not use monotask boards. Use /mastermind-issues for org issues."; exit 1; }
 ```
 
 ---

package/.claude/skills/mastermind-threads/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-threads
 description: Mastermind threads — list, view, and create conversation threads within an org. Threads are human-or-agent discussions attached to issues, goals, or the org itself. Reads from -threads.jsonl org state files.
 type: domain-skill
 default_mode: auto
+pick: low
 ---
 
 # Mastermind Threads
 
-This skill is invoked by `mastermind:threads` or directly via `/mastermind:threads`.
+This skill is invoked by `mastermind:threads` or directly via `/mastermind-threads`.
 
 ---
 
@@ -54,7 +55,7 @@ echo "────────────────────────
 if [ ! -f "$threadsFile" ] || [ ! -s "$threadsFile" ]; then
   echo "  No threads found."
   echo ""
-  echo "  Create: /mastermind:threads --org $org_name --action create --message 'Hello team'"
+  echo "  Create: /mastermind-threads --org $org_name --action create --message 'Hello team'"
   exit 0
 fi
 
@@ -94,7 +95,7 @@ else:
         created = (t.get("createdAt") or "-")[:10]
         print(f"  {tid:<28} {subj:<32} {author:<20} {msgs:<6} {created}")
 
-print(f"\n  {len(threads)} thread(s). View: /mastermind:threads --org <org> --action view --thread-id <id>")
+print(f"\n  {len(threads)} thread(s). View: /mastermind-threads --org <org> --action view --thread-id <id>")
 PYEOF
 ```
 

package/.claude/skills/mastermind-tree-control/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-tree-control
 description: Mastermind tree-control — pause, hold, release, or preview recovery for an issue/task tree in an org. Lets board members stop runaway loops, hold trees during review, and resume work when ready. Mirrors Paperclip's issue-tree-control API.
 type: domain-skill
 default_mode: confirm
+pick: low
 ---
 
 # Mastermind Tree Control
 
-This skill is invoked by `mastermind:tree-control` or directly via `/mastermind:tree-control`.
+This skill is invoked by `mastermind:tree-control` or directly via `/mastermind-tree-control`.
 
 ---
 
@@ -94,8 +95,8 @@ print()
 print(f"  Active issues in subtree: {len(active)}")
 print(f"  Holding root would affect all {len(active)} active issue(s).")
 print()
-print("  To hold: /mastermind:tree-control --org <org> --action hold --issue-id <id> --reason 'Review needed'")
-print("  To cancel: /mastermind:tree-control --org <org> --action cancel --issue-id <id> --reason 'No longer needed'")
+print("  To hold: /mastermind-tree-control --org <org> --action hold --issue-id <id> --reason 'Review needed'")
+print("  To cancel: /mastermind-tree-control --org <org> --action cancel --issue-id <id> --reason 'No longer needed'")
 PYEOF
 ```
 
@@ -140,7 +141,7 @@ print(f"  Reason: {reason}")
 print(f"  Applied: {ts}")
 print()
 print("  Active agents should detect the hold and pause execution.")
-print(f"  To release: /mastermind:tree-control --org <org> --action release --issue-id {rootId}")
+print(f"  To release: /mastermind-tree-control --org <org> --action release --issue-id {rootId}")
 PYEOF
 ```
 

package/.claude/skills/mastermind-workspace-detail/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-workspace-detail
 description: Mastermind workspace-detail — deep per-execution-workspace inspection and runtime control. Manage services, provision/teardown/cleanup commands, linked issues and routines, runtime logs, and configuration for a single execution workspace.
 type: domain-skill
 default_mode: confirm
+pick: low
 ---
 
 # Mastermind Workspace Detail
 
-This skill is invoked by `mastermind:workspace-detail` or directly via `/mastermind:workspace-detail`.
+This skill is invoked by `mastermind:workspace-detail` or directly via `/mastermind-workspace-detail`.
 
 ---
 
@@ -54,7 +55,7 @@ orgFile=".monomind/orgs/${org_name}.json"
 [ ! -f "$orgFile" ] && { echo "ERROR: Org '${org_name}' not found."; exit 1; }
 
 wsFile=".monomind/orgs/${org_name}-workspaces.json"
-[ ! -f "$wsFile" ] && { echo "ERROR: No workspaces file for org '$org_name'. Create workspaces via /mastermind:workspaces."; exit 1; }
+[ ! -f "$wsFile" ] && { echo "ERROR: No workspaces file for org '$org_name'. Create workspaces via /mastermind-workspaces."; exit 1; }
 
 wsDef=$(jq -r --arg id "$workspace_id" '(.workspaces // [])[] | select(.id == $id)' "$wsFile")
 [ -z "$wsDef" ] && { echo "ERROR: Workspace '$workspace_id' not found in org '$org_name'."; exit 1; }
@@ -256,7 +257,7 @@ else
 fi
 
 echo ""
-echo "To run a routine with workspace vars: /mastermind:routine-detail --org $org_name --routine-id <id> --action variables"
+echo "To run a routine with workspace vars: /mastermind-routine-detail --org $org_name --routine-id <id> --action variables"
 ```
 
 ### config

package/.claude/skills/mastermind-workspaces/SKILL.md

@@ -3,11 +3,12 @@ name: mastermind-workspaces
 description: Mastermind workspaces — manage isolated git worktree workspaces per project. List, attach, detach, stop, and prune workspaces. Grouped by project with running service counts and agent assignments.
 type: domain-skill
 default_mode: confirm
+pick: low
 ---
 
 # Mastermind Workspaces
 
-This skill is invoked by `mastermind:workspaces` or directly via `/mastermind:workspaces`.
+This skill is invoked by `mastermind:workspaces` or directly via `/mastermind-workspaces`.
 
 ---
 

package/.claude/skills/monoswarm/SKILL.md

@@ -99,11 +99,13 @@ Common assignments (see `doc/concepts/monoswarm.md` for the full routing table):
 | Task | Recommended agents |
 |---|---|
 | Bug fix | coordinator, researcher, coder, tester |
-| Feature | coordinator, architect, coder, tester, reviewer |
-| Refactor | coordinator, architect, coder, reviewer |
-| Performance | coordinator, performance-engineer, coder |
-| Security | coordinator, security-architect, security-auditor |
-| Docs | researcher, documenter |
+| Feature | coordinator, Software Architect, coder, tester, reviewer |
+| Refactor | coordinator, Software Architect, coder, reviewer |
+| Performance | coordinator, Performance Benchmarker, coder |
+| Security | coordinator, Security Engineer, reviewer |
+| Docs | researcher, Technical Writer |
+
+These are real agent names (Task `subagent_type`). For a task-specific pick use the prompt's `[PICK]` line or `mcp__monomind__pick`.
 
 ## Workflow Patterns
 

package/.claude/skills/specialagent/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: specialagent
-description: Find the single best specialized agent from the verified ~98-agent roster using two-stage LLM domain→agent selection
+description: Find the single best specialized agent for a task — the [PICK] line or the shared pick index (mcp__monomind__pick / monomind pick), with a two-stage domain→agent fallback over the installed agents
 version: 2.0.0
 triggers:
   - /specialagent
@@ -18,87 +18,68 @@ triggers:
   - which swarm agent
 tools:
   - Bash
+pick: low
 ---
 
-# /specialagent — Two-Stage LLM Agent Selection
+# /specialagent — Best Agent for a Task
 
-Finds the best agent using a lightweight two-stage LLM approach: first pick the domain, then pick the agent within that domain. Only names are passed at each stage — no descriptions, no keyword dumps.
+Finds the best installed agent for a task. Never recommend from memory or from a hardcoded roster: the agent set differs per install, and a `subagent_type` that is not installed fails at spawn time.
 
-## How It Works
+## Step 1: Use the pick index
 
-```
-Stage 1: Give LLM domain names → LLM picks best domain
-Stage 2: Give LLM agent names in that domain → LLM picks best agent
-```
+In this order, stop at the first that answers:
 
-## Stage 1: Domain Selection
+1. **`[PICK]` line** — if the prompt carries `[PICK] agent: <name> · skill: <invoke>`, recommend that agent unless it is clearly wrong for the task.
+2. **`mcp__monomind__pick({ task: "<task>", kind: "agents", top: 5 })`** — choose the best fit among `agents.ranked[]` (prefer the top entry unless another is clearly more specific). Each `name` is a spawnable `subagent_type`.
+3. **Local CLI** — `monomind pick -t "<task>" --top 5 --json | jq -r '.agents.ranked[].name'` (local binary only, never npx; see `mastermind-agent-select/SKILL.md` for the version-checked `mmpick` helper).
 
-The verified roster is ~98 agent definitions (97 registered in `packages/@monomind/cli/.monomind/registry.json`). Domains and representative agents from each:
+## Step 2 (fallback): Two-stage domain → agent selection
 
-| Domain | Representative agents (NOT exhaustive — run `ls .claude/agents/` for the full set) |
-|---|---|
-| development | coder · Backend Architect · Frontend Developer · mobile-dev · Mobile App Builder · Rapid Prototyper · Software Architect · Senior Developer · AI Engineer · Data Engineer · Database Optimizer · AI Data Remediation Engineer · LSP/Index Engineer · Embedded Firmware Engineer · Solidity Smart Contract Engineer · WeChat Mini Program Developer · Feishu Integration Developer · Model QA Specialist |
-| testing | tester · tdd-london-swarm · production-validator · API Tester · Accessibility Auditor · Evidence Collector · Performance Benchmarker · Test Results Analyzer · Tool Evaluator · Workflow Optimizer · Code Reviewer |
-| security | Security Engineer · Compliance Auditor · Blockchain Security Auditor · Threat Detection Engineer · Agentic Identity & Trust Architect · Identity Graph Operator · ZK Steward |
-| devops | DevOps Automator · SRE (Site Reliability Engineer) · Git Workflow Master · Incident Response Commander |
-| architecture | system-architect · Software Architect · Backend Architect · Workflow Architect · Autonomous Optimization Architect · Automation Governance Architect · v1-integration-architect |
-| research | researcher · Technical Writer · Developer Advocate · goal-planner |
-| marketing | Competitive Content Strategist · CRO Specialist · Email Marketing Specialist · Launch Strategist · Pricing Strategist |
-| design | Monodesign · Cultural Intelligence Strategist |
-| github | pr-manager · issue-tracker · release-manager · repo-architect · code-review-swarm · multi-repo-swarm · sync-coordinator · workflow-automation · project-board-sync |
-| swarm / consensus | mesh-coordinator · collective-intelligence-coordinator · scout-explorer · swarm-memory-manager · worker-specialist · coordinator · planner · smart-agent · swarm-init · quorum-manager |
-| optimization | Benchmark Suite · Load Balancing Coordinator · Performance Monitor · Resource Allocator · Topology Optimizer |
-| specialized | MCP Builder · Document Generator · Agents Orchestrator · dashboard-verifier |
+Only when Step 1 returned nothing. Only names are passed at each stage — no descriptions, no keyword dumps.
 
-**The filesystem is authoritative** — run `ls .claude/agents/` (and `ls packages/@monomind/cli/.claude/agents/`) for the current full list; agents are added over time. Only recommend agents that actually exist there.
+```
+Stage 1: Give LLM the category names → LLM picks the best category
+Stage 2: Give LLM the agent names installed in that category → LLM picks the best agent
+```
 
-**Stage 1 prompt to yourself:**
-> "Given the task: `<task>` — which single domain from this list best fits: development, testing, security, devops, architecture, research, marketing, design, github, swarm, optimization, specialized? Answer with just the domain name."
+Categories are the `.claude/agents/` folder names: `architecture consensus core design engineering github goal marketing monoswarm optimization specialists specialized templates testing`.
+
+List the installed agent names in a category from the registry (or the definitions on disk):
 
-## Stage 2: Agent Selection Per Domain
+```bash
+CAT="engineering"
+jq -r --arg c "$CAT" '.agents[] | select(.category == $c and .deprecated != true) | .name' .monomind/registry.json 2>/dev/null \
+  || grep -h -m1 "^name:" packages/@monomind/cli/.claude/agents/$CAT/*.md .claude/agents/$CAT/*.md 2>/dev/null | sed 's/^name: *//'
+```
 
-From the chosen domain's agents (table above), pick the best fit. **Before recommending, verify the name exists** as a `.md` file under `.claude/agents/` (or `packages/@monomind/cli/.claude/agents/`). If the name you want is not on disk, fall back to another agent in the domain or to a core agent (`coder`, `reviewer`, `tester`, `researcher`, `planner`).
+**Stage 1 prompt to yourself:**
+> "Given the task: `<task>` — which single category from this list best fits: `<category names>`? Answer with just the category name."
 
 **Stage 2 prompt to yourself:**
-> "Given the task: `<task>` — which single agent from this domain's list is the best fit: `<comma-separated agent names for the selected domain>`? Answer with just the agent name."
+> "Given the task: `<task>` — which single agent from this category is the best fit: `<agent names listed above>`? Answer with just the agent name."
 
-## Execution Steps
-
-1. Read the user's task
-2. **Stage 1**: Internally reason through the domain list → select one domain
-3. **Stage 2**: Internally reason through agent names in that domain → select one agent
-4. Verify the agent exists: `ls .claude/agents/**/<slug>.md` (names double as slugs; confirm via the frontmatter `name:` field)
-5. Output the recommendation
+If both stages come up empty, fall back to a core agent: `coder`, `reviewer`, `tester`, `researcher`, `planner`.
 
 ## Slug Mapping
 
-Agent names double as their `subagent_type` slug — the slug is the agent's frontmatter `name` (or the filename without `.md`). Resolve a name to its slug from the definition on disk:
-
-```bash
-# list every available agent definition
-ls .claude/agents/ packages/@monomind/cli/.claude/agents/
-# read a specific agent's registered name
-grep -m1 "^name:" .claude/agents/**/<file>.md
-```
-
-If unsure, pass the agent name as the slug. The authoritative set is whatever currently exists in those directories (~98 definitions).
+The `subagent_type` is the agent's frontmatter `name:` — what the pick index returns as `name`, and what the registry lists as `.name`.
 
 ## Output Format
 
 ```
 TASK: <one-line task summary>
 
-DOMAIN: <selected domain>
+PICKED BY: <[PICK] line | mcp__monomind__pick | monomind pick | category fallback: <category>>
 
 RECOMMENDED AGENT: <Agent Name>
-Invoke: Task({ subagent_type: "<slug>", prompt: "..." })
+Invoke: Task({ subagent_type: "<agent name>", prompt: "..." })
 ```
 
 Then ask: "Should I spawn this agent now?"
 
 ## Rules
 
-1. Only pass names at each stage — no descriptions, no keyword dumps, no scoring tables
-2. Pick exactly one domain, then exactly one agent
+1. Prefer the pick index; in the fallback, only pass names at each stage — no descriptions, no keyword dumps, no scoring tables
+2. Recommend exactly one agent, and only a name that is installed
 3. For tasks that clearly need a specialized tool (e.g. accessibility audits → Accessibility Auditor, not tester), prefer the more specific agent
 4. Never recommend a generic role (coder, tester) when a specialized agent in the right domain exists