opencode-onboard 0.2.13 → 0.3.1

package/content/.agents/agents/back-engineer.md

@@ -33,15 +33,17 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
-1. Read the task and identify domain and platform
-2. Scan `.agents/skills/` for available skills
-3. Read each `SKILL.md` description to assess relevance
-4. Load and follow any skill that applies, even partial match warrants loading
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, read the task and identify domain and platform
+3. Scan `.agents/skills/` for available skills
+4. Read each `SKILL.md` description to assess relevance
+5. Load and follow any skill that applies, even partial match warrants loading
 
 Rules:
 - Never implement directly if a skill applies
 - Follow skill instructions exactly, do not partially apply them
 - If two skills apply, follow both, resolve conflicts by asking the lead
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Responsibilities
 
@@ -66,8 +68,9 @@ Rules:
 ## Workflow
 
 When spawned by the lead:
+0. Read ALL skills listed in the spawn prompt FIRST. Do not proceed until every listed SKILL.md has been read. Reply to lead with `team_message` confirming which skills were loaded.
 1. For each assigned task: call `team_claim task_id:<id>` before starting
-2. Implement the task
+2. Implement the task following loaded skill rules
 3. Call `team_tasks_complete task_id:<id>` after finishing
 4. When all tasks are done or blocked, send results to lead via `team_message`
 

package/content/.agents/agents/devops-manager.md

@@ -33,6 +33,9 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, identify the platform from URLs or context
+
 Examples of intent → skill mapping:
 - URL contains `dev.azure.com` or `visualstudio.com` → look for `ob-userstory-az` or `ob-pullrequest-az`
 - URL contains `github.com` → look for `ob-userstory-gh` or `ob-pullrequest-gh`
@@ -43,6 +46,7 @@ Rules:
 - Never interact with a platform without loading the matching skill first
 - Follow skill instructions exactly, do not partially apply them
 - If no skill exists for the platform, report it as a blocker rather than improvising
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Two Modes
 

package/content/.agents/agents/front-engineer.md

@@ -33,15 +33,17 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
-1. Read the task and identify domain and platform
-2. Scan `.agents/skills/` for available skills
-3. Read each `SKILL.md` description to assess relevance
-4. Load and follow any skill that applies, even partial match warrants loading
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, read the task and identify domain and platform
+3. Scan `.agents/skills/` for available skills
+4. Read each `SKILL.md` description to assess relevance
+5. Load and follow any skill that applies, even partial match warrants loading
 
 Rules:
 - Never implement directly if a skill applies
 - Follow skill instructions exactly, do not partially apply them
 - If two skills apply, follow both, resolve conflicts by asking the lead
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Responsibilities
 
@@ -65,8 +67,9 @@ Rules:
 ## Workflow
 
 When spawned by the lead:
+0. Read ALL skills listed in the spawn prompt FIRST. Do not proceed until every listed SKILL.md has been read. Reply to lead with `team_message` confirming which skills were loaded.
 1. For each assigned task: call `team_claim task_id:<id>` before starting
-2. Implement the task
+2. Implement the task following loaded skill rules
 3. Call `team_tasks_complete task_id:<id>` after finishing
 4. When all tasks are done or blocked, send results to lead via `team_message`
 

package/content/.agents/agents/infra-engineer.md

@@ -33,15 +33,17 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
-1. Read the task and identify domain and platform
-2. Scan `.agents/skills/` for available skills
-3. Read each `SKILL.md` description to assess relevance
-4. Load and follow any skill that applies, even partial match warrants loading
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, read the task and identify domain and platform
+3. Scan `.agents/skills/` for available skills
+4. Read each `SKILL.md` description to assess relevance
+5. Load and follow any skill that applies, even partial match warrants loading
 
 Rules:
 - Never implement directly if a skill applies
 - Follow skill instructions exactly, do not partially apply them
 - If two skills apply, follow both, resolve conflicts by asking the lead
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Responsibilities
 
@@ -62,6 +64,15 @@ Rules:
 - Do not force push
 - Report blockers immediately rather than working around them
 
+## Workflow
+
+When spawned by the lead:
+0. Read ALL skills listed in the spawn prompt FIRST. Do not proceed until every listed SKILL.md has been read. Reply to lead with `team_message` confirming which skills were loaded.
+1. For each assigned task: call `team_claim task_id:<id>` before starting
+2. Implement the task following loaded skill rules
+3. Call `team_tasks_complete task_id:<id>` after finishing
+4. When all tasks are done or blocked, send results to lead via `team_message`
+
 ## Output Format
 
 ```

package/content/.agents/agents/quality-engineer.md

@@ -33,15 +33,17 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
-1. Read the task and identify domain and platform
-2. Scan `.agents/skills/` for available skills
-3. Read each `SKILL.md` description to assess relevance
-4. Load and follow any skill that applies, even partial match warrants loading
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, read the task and identify domain and platform
+3. Scan `.agents/skills/` for available skills
+4. Read each `SKILL.md` description to assess relevance
+5. Load and follow any skill that applies, even partial match warrants loading
 
 Rules:
 - Never implement directly if a skill applies
 - Follow skill instructions exactly, do not partially apply them
 - If two skills apply, follow both, resolve conflicts by asking the lead
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Responsibilities
 
@@ -64,8 +66,9 @@ Rules:
 
 When spawned by the lead:
 1. Read the task list and context files provided in the spawn prompt
-2. Run tests, build, lint, and verify acceptance criteria
-3. When done, send results to lead via `team_message`
+2. If no board task is assigned, do not block on team_claim. Run verification directly from the spawn prompt scope.
+3. Run tests, build, lint, and verify acceptance criteria
+4. When done, send results to lead via `team_message`
 
 ## Output Format
 

package/content/.agents/agents/security-auditor.md

@@ -31,15 +31,17 @@ If `rtk` is not available, report it as a blocker. Do not run commands without i
 
 Skills are located in `.agents/skills/`. Detect and use relevant skills automatically, the user will never tell you which skill to use.
 
-1. Read the task and identify domain and platform
-2. Scan `.agents/skills/` for available skills
-3. Read each `SKILL.md` description to assess relevance
-4. Load and follow any skill that applies, even partial match warrants loading
+1. If the spawn prompt lists specific skills to load, read those `SKILL.md` files FIRST before any implementation
+2. Additionally, read the task and identify domain and platform
+3. Scan `.agents/skills/` for available skills
+4. Read each `SKILL.md` description to assess relevance
+5. Load and follow any skill that applies, even partial match warrants loading
 
 Rules:
 - Never implement directly if a skill applies
 - Follow skill instructions exactly, do not partially apply them
 - If two skills apply, follow both, resolve conflicts by asking the lead
+- Skills listed in the spawn prompt are MANDATORY, not optional
 
 ## Responsibilities
 

package/content/.agents/skills/ob-pullrequest-gh/SKILL.md

@@ -65,14 +65,19 @@ rtk gh pr create \
 
 ### Step 5: Post screenshot comment
 
-Build raw URL for each image:
+Resolve commit SHA (the commit that includes screenshots):
+```bash
+rtk git rev-parse HEAD
+```
+
+Build blob URL for each image (preferred, stable in PR discussion):
 ```
-https://raw.githubusercontent.com/{owner}/{repo}/feature/{slug}/openspec/changes/{change}/images/{file}.png
+https://github.com/{owner}/{repo}/blob/{sha}/openspec/changes/{change}/images/{file}.png
 ```
 
 Post comment:
 ```bash
-rtk gh pr comment {pr-number} --body $'## Screenshots\n\n![{feature}]({raw-url})'
+rtk gh pr comment {pr-number} --repo {owner}/{repo} --body $'## Screenshots\n\n![{feature}]({blob-url})'
 ```
 
 ---

package/content/.agents/skills/ob-userstory-gh/SKILL.md

@@ -80,16 +80,16 @@ rtk gh issue edit <number> --repo {owner}/{repo} --add-label "in-progress"
 
 ## Screenshot / Image Strategy
 
-**Never embed images as attachments.** Save to openspec change folder and reference via raw GitHub URL.
+**Never embed images as attachments.** Save to openspec change folder and reference via GitHub blob URL pinned to commit SHA.
 
 ### Save location
 ```
 openspec/changes/{change-name}/images/{screenshot}.png
 ```
 
-### Raw URL format (renders inline in PR comments)
+### Blob URL format (preferred)
 ```
-https://raw.githubusercontent.com/{owner}/{repo}/{branch}/openspec/changes/{change}/images/{file}.png
+https://github.com/{owner}/{repo}/blob/{sha}/openspec/changes/{change}/images/{file}.png
 ```
 
 ---
@@ -103,8 +103,8 @@ https://github.com/{owner}/{repo}/issues/{number}
 # PR
 https://github.com/{owner}/{repo}/pull/{number}
 
-# Raw file
-https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path}
+# Blob file
+https://github.com/{owner}/{repo}/blob/{sha}/{path}
 ```
 
 ---

package/content/.opencode/commands/opsx-apply.md

@@ -86,7 +86,14 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
       DO NOT call team_claim yourself, only agents claim tasks.
       DO NOT proceed to 6d until team_tasks_add succeeds.
 
-   **Step 6d.** Spawn all needed specialists, then kick them off in parallel.
+   **Step 6d.** Discover relevant skills, then spawn specialists.
+
+      Before spawning, scan `.agents/skills/` and read each `SKILL.md` description line.
+      Match skills to agents by domain:
+      - front-engineer: UI, components, framework skills (e.g. next-best-practices, browser-automation)
+      - back-engineer: API, data, service skills
+      - infra-engineer: cloud, pipeline, deployment skills
+      - quality-engineer: testing, coverage skills
 
       Each team_spawn MUST include the agent field (required, causes NOT NULL error if omitted).
 
@@ -97,8 +104,9 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
       4. The 6 OpenCode tools they have available (these are OpenCode tools, NOT shell commands, call them directly as tools, never via bash):
          team_claim, team_tasks_complete, team_tasks_list, team_tasks_add, team_message, team_broadcast
       5. How to proceed: call team_claim tool with the task_id to claim a task before starting it, call team_tasks_complete tool after finishing it, repeat until all their tasks are done, then call team_message tool to notify lead with results or blockers
+      6. Which skills to load: list the skill names and paths they MUST read before implementing. Example: "Before starting, read `.agents/skills/next-best-practices/SKILL.md` and follow its rules for all Next.js code."
 
-      Keep spawn prompts under 500 tokens. Do not describe team internals or how ensemble works.
+      Keep spawn prompts under 600 tokens. Do not describe team internals or how ensemble works.
       Only spawn agents whose tasks are actually needed by this change. Skip agents with no tasks.
 
       First spawn all agents (wait for each team_spawn to confirm before the next):
@@ -113,9 +121,9 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
 
       Then immediately send each spawned agent a start message to kick them off:
       ```
-      team_message to:"back" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"front" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"infra" text:"Start now. Claim your first task with team_claim and begin implementing."
+      team_message to:"back" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"front" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"infra" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
       ```
 
    **Step 6e.** After sending start messages, tell the user what is running, then STOP and wait.
@@ -131,7 +139,7 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
 
    Spawn quality engineer with worktree:false (read-only, no file edits):
    ```
-   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<task list, context summary, run tests + build + lint + verify acceptance criteria, send results to lead when done>"
+   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<verification scope, context summary, run tests + build + lint + verify acceptance criteria, no task claiming required in this phase, send results to lead when done>"
    ```
    Wait for message → team_results → fix blockers → team_shutdown (no team_merge needed, worktree:false)
 

package/content/.opencode/plugins/session-log.js

@@ -2,7 +2,8 @@ import fs from "node:fs"
 import path from "node:path"
 
 const LOG_FILE = ".agents/session-log.json"
-const SPAWN_MATCH_WINDOW_MS = 15000
+const SPAWN_MATCH_WINDOW_MS = 30000
+const UNMATCHED_SESSION_WARN_MS = 30000
 const DEBUG = process.env.SESSION_LOG_DEBUG === "true"
 
 // Per-session state
@@ -12,8 +13,12 @@ const sessionState = new Map()
 const leadTeamBySession = new Map()
 
 // Pending spawn records waiting for a session.created match
+// Each entry: { leadSessionId, at, member, agentType, teamName, spawnedSessionId, intendedSkills }
 const pendingSpawns = []
 
+// spawnedSessionId -> pending spawn (direct correlation fast path)
+const pendingSpawnBySessionId = new Map()
+
 // Team -> completed session snapshots
 const completedByTeam = new Map()
 
@@ -47,6 +52,11 @@ function resolveAgentName(session) {
   return "lead"
 }
 
+function resolveModel(session) {
+  // Try common field names used by OpenCode session objects
+  return session?.model || session?.modelId || session?.model_id || null
+}
+
 function addSkillToState(state, skillName) {
   if (!skillName || !state) return false
   if (!state.skills) state.skills = new Set()
@@ -55,6 +65,37 @@ function addSkillToState(state, skillName) {
   return true
 }
 
+// Extract skill names hinted in a spawn prompt by scanning for known skill-like tokens.
+// Matches strings that look like skill directory names (kebab-case words after "skill" keyword
+// or adjacent to known skill name patterns).
+function extractIntendedSkills(prompt) {
+  if (!prompt || typeof prompt !== "string") return []
+  const skills = new Set()
+
+  // Match explicit skill name mentions: e.g. "ob-pullrequest-gh", "browser-automation"
+  const kebabPattern = /\b([a-z][a-z0-9]*(?:-[a-z0-9]+){1,})\b/g
+  let m
+  while ((m = kebabPattern.exec(prompt)) !== null) {
+    const candidate = m[1]
+    // Filter to plausible skill names: 2+ segments, known prefixes or suffixes
+    if (
+      candidate.startsWith("ob-") ||
+      candidate.startsWith("openspec-") ||
+      candidate.startsWith("browser-") ||
+      candidate.endsWith("-gh") ||
+      candidate.endsWith("-az") ||
+      candidate.endsWith("-automation") ||
+      candidate.endsWith("-change") ||
+      candidate.endsWith("-engineer") ||
+      candidate.endsWith("-manager") ||
+      candidate.endsWith("-auditor")
+    ) {
+      skills.add(candidate)
+    }
+  }
+  return Array.from(skills).sort()
+}
+
 function toNum(v) {
   if (typeof v === "number" && Number.isFinite(v)) return v
   if (typeof v === "string" && v.trim() !== "" && !Number.isNaN(Number(v))) return Number(v)
@@ -133,7 +174,9 @@ function buildCompletedSnapshot(state, sessionId) {
     member: state.member || null,
     agentType: state.agentType || null,
     team: state.teamName || null,
+    model: state.model || null,
     skills: Array.from(state.skills || []).sort(),
+    intendedSkills: state.intendedSkills || [],
     usage: usagePayload(state),
     filesEdited: state.editCount || 0,
   }
@@ -144,14 +187,21 @@ function buildTeamSkillsSummary(teamName) {
   const byAgent = {}
   for (const row of rows) {
     const key = row.member || row.agent || "unknown"
-    if (!byAgent[key]) byAgent[key] = { agentType: row.agentType || null, skills: new Set() }
+    if (!byAgent[key]) byAgent[key] = { agentType: row.agentType || null, model: row.model || null, skills: new Set(), intendedSkills: new Set() }
     for (const s of row.skills || []) byAgent[key].skills.add(s)
+    for (const s of row.intendedSkills || []) byAgent[key].intendedSkills.add(s)
   }
   const out = {}
   for (const [k, v] of Object.entries(byAgent)) {
+    const skills = Array.from(v.skills).sort()
+    const intendedSkills = Array.from(v.intendedSkills).sort()
+    const missingSkills = intendedSkills.filter(s => !v.skills.has(s))
     out[k] = {
       agentType: v.agentType,
-      skills: Array.from(v.skills).sort(),
+      model: v.model,
+      skills,
+      intendedSkills,
+      missingSkills: missingSkills.length > 0 ? missingSkills : undefined,
     }
   }
   return out
@@ -163,21 +213,52 @@ function trackCompletedByTeam(snapshot) {
   completedByTeam.get(snapshot.team).push(snapshot)
 }
 
-function enqueuePendingSpawn(leadSessionId, args) {
-  pendingSpawns.push({
+function enqueuePendingSpawn(leadSessionId, args, spawnOutput) {
+  // Try to extract spawnedSessionId from team_spawn output for direct correlation
+  const spawnedSessionId =
+    spawnOutput?.sessionId ||
+    spawnOutput?.session_id ||
+    spawnOutput?.id ||
+    spawnOutput?.data?.sessionId ||
+    spawnOutput?.data?.session_id ||
+    spawnOutput?.data?.id ||
+    null
+
+  const record = {
     leadSessionId,
     at: nowMs(),
     member: args?.name || null,
     agentType: args?.agent || null,
     teamName: leadTeamBySession.get(leadSessionId) || null,
-  })
+    spawnedSessionId,
+    intendedSkills: extractIntendedSkills(args?.prompt),
+  }
+
+  pendingSpawns.push(record)
+
+  if (spawnedSessionId) {
+    pendingSpawnBySessionId.set(spawnedSessionId, record)
+  }
 }
 
-function matchPendingSpawn() {
+function matchPendingSpawn(sessionId) {
   const now = nowMs()
-  // Drop expired pending spawns first
+
+  // Fast path: direct session ID correlation from team_spawn output
+  if (sessionId && pendingSpawnBySessionId.has(sessionId)) {
+    const record = pendingSpawnBySessionId.get(sessionId)
+    pendingSpawnBySessionId.delete(sessionId)
+    const idx = pendingSpawns.indexOf(record)
+    if (idx !== -1) pendingSpawns.splice(idx, 1)
+    return record
+  }
+
+  // Fallback: time-window heuristic (drop expired first)
   for (let i = pendingSpawns.length - 1; i >= 0; i--) {
-    if (now - pendingSpawns[i].at > SPAWN_MATCH_WINDOW_MS) pendingSpawns.splice(i, 1)
+    if (now - pendingSpawns[i].at > SPAWN_MATCH_WINDOW_MS) {
+      pendingSpawnBySessionId.delete(pendingSpawns[i].spawnedSessionId)
+      pendingSpawns.splice(i, 1)
+    }
   }
   if (pendingSpawns.length === 0) return null
   return pendingSpawns.shift()
@@ -210,13 +291,16 @@ export const SessionLogPlugin = async ({ client, directory }) => {
           const res = await client.session.get({ path: { id: sessionId } })
           const session = res?.data
           const fallbackAgent = resolveAgentName(session)
-          const spawnMatch = matchPendingSpawn()
+          const model = resolveModel(session)
+          const spawnMatch = matchPendingSpawn(sessionId)
 
           const state = {
             agentName: spawnMatch?.member || fallbackAgent,
             member: spawnMatch?.member || null,
             agentType: spawnMatch?.agentType || null,
             teamName: spawnMatch?.teamName || null,
+            model,
+            intendedSkills: spawnMatch?.intendedSkills || [],
             editCount: 0,
             skills: new Set(),
             startedAtMs: nowMs(),
@@ -226,18 +310,56 @@ export const SessionLogPlugin = async ({ client, directory }) => {
             reportedInputTokens: 0,
             reportedOutputTokens: 0,
             reportedTotalTokens: 0,
+            // Track whether this session was matched to a spawn record
+            spawnMatched: !!spawnMatch,
           }
 
           sessionState.set(sessionId, state)
-          appendEntry(directory, {
+
+          const startedEntry = {
             ts: ts(),
             agent: state.agentName,
             member: state.member,
             agentType: state.agentType,
             team: state.teamName,
+            model: state.model,
             action: "started",
             sessionId,
-          })
+          }
+          appendEntry(directory, startedEntry)
+
+          // Emit explicit teammate-registered entry when a spawn is matched
+          if (spawnMatch) {
+            appendEntry(directory, {
+              ts: ts(),
+              agent: state.agentName,
+              member: state.member,
+              agentType: state.agentType,
+              team: state.teamName,
+              model: state.model,
+              intendedSkills: state.intendedSkills,
+              action: "teammate-registered",
+              sessionId,
+              correlationMethod: spawnMatch.spawnedSessionId === sessionId ? "direct" : "time-window",
+            })
+          } else {
+            // No spawn match — schedule an unmatched-session warning
+            const capturedSessionId = sessionId
+            setTimeout(() => {
+              const s = sessionState.get(capturedSessionId)
+              if (!s || s.spawnMatched) return
+              appendEntry(directory, {
+                ts: ts(),
+                agent: s.agentName,
+                member: s.member,
+                team: s.teamName,
+                model: s.model,
+                action: "unmatched-session",
+                sessionId: capturedSessionId,
+                warning: "Session started with no matching team_spawn record. Agent identity may be inaccurate.",
+              })
+            }, UNMATCHED_SESSION_WARN_MS)
+          }
         }
 
         if (event?.type === "file.edited") {
@@ -261,9 +383,11 @@ export const SessionLogPlugin = async ({ client, directory }) => {
             member: state.member,
             agentType: state.agentType,
             team: state.teamName,
+            model: state.model,
             action: "completed",
             filesEdited: state.editCount,
             skills,
+            intendedSkills: state.intendedSkills,
             usage,
           })
 
@@ -321,6 +445,7 @@ export const SessionLogPlugin = async ({ client, directory }) => {
               member: state.member,
               agentType: state.agentType,
               team: state.teamName,
+              model: state.model,
               action: "skill-loaded",
               skill: skillName,
               source: "skill-tool",
@@ -343,6 +468,7 @@ export const SessionLogPlugin = async ({ client, directory }) => {
                 member: state.member,
                 agentType: state.agentType,
                 team: state.teamName,
+                model: state.model,
                 action: "skill-loaded",
                 skill: skillName,
                 source: "read-skill-file",
@@ -362,6 +488,7 @@ export const SessionLogPlugin = async ({ client, directory }) => {
           member: state.member,
           agentType: state.agentType,
           team: state.teamName,
+          model: state.model,
           ...ensembleHandler(args),
         }
         appendEntry(directory, entry)
@@ -372,7 +499,7 @@ export const SessionLogPlugin = async ({ client, directory }) => {
         }
 
         if (tool === "team_spawn") {
-          enqueuePendingSpawn(sessionId, args)
+          enqueuePendingSpawn(sessionId, args, output)
         }
 
         if (tool === "team_cleanup") {
@@ -380,6 +507,7 @@ export const SessionLogPlugin = async ({ client, directory }) => {
           appendEntry(directory, {
             ts: ts(),
             agent: state.agentName,
+            model: state.model,
             action: "team-skills-summary",
             team: teamName || null,
             byAgent: teamName ? buildTeamSkillsSummary(teamName) : {},

package/content/.opencode/skills/openspec-apply-change/SKILL.md

@@ -92,7 +92,14 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
       DO NOT call team_claim yourself, only agents claim tasks.
       DO NOT proceed to 6d until team_tasks_add succeeds.
 
-   **Step 6d.** Spawn all needed specialists, then kick them off in parallel.
+   **Step 6d.** Discover relevant skills, then spawn specialists.
+
+      Before spawning, scan `.agents/skills/` and read each `SKILL.md` description line.
+      Match skills to agents by domain:
+      - front-engineer: UI, components, framework skills (e.g. next-best-practices, browser-automation)
+      - back-engineer: API, data, service skills
+      - infra-engineer: cloud, pipeline, deployment skills
+      - quality-engineer: testing, coverage skills
 
       Each team_spawn MUST include the agent field (required, causes NOT NULL error if omitted).
 
@@ -103,8 +110,9 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
       4. The 6 OpenCode tools they have available (these are OpenCode tools, NOT shell commands, call them directly as tools, never via bash):
          team_claim, team_tasks_complete, team_tasks_list, team_tasks_add, team_message, team_broadcast
       5. How to proceed: call team_claim tool with the task_id to claim a task before starting it, call team_tasks_complete tool after finishing it, repeat until all their tasks are done, then call team_message tool to notify lead with results or blockers
+      6. Which skills to load: list the skill names and paths they MUST read before implementing. Example: "Before starting, read `.agents/skills/next-best-practices/SKILL.md` and follow its rules for all Next.js code."
 
-      Keep spawn prompts under 500 tokens. Do not describe team internals or how ensemble works.
+      Keep spawn prompts under 600 tokens. Do not describe team internals or how ensemble works.
       Only spawn agents whose tasks are actually needed by this change. Skip agents with no tasks.
 
       First spawn all agents (wait for each team_spawn to confirm before the next):
@@ -119,9 +127,9 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
 
       Then immediately send each spawned agent a start message to kick them off:
       ```
-      team_message to:"back" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"front" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"infra" text:"Start now. Claim your first task with team_claim and begin implementing."
+      team_message to:"back" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"front" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"infra" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
       ```
 
    **Step 6e.** After sending start messages, tell the user what is running, then STOP and wait.
@@ -137,7 +145,7 @@ Implement tasks from an OpenSpec change using the ensemble agent team.
 
    Spawn quality engineer with worktree:false (read-only, no file edits):
    ```
-   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<task list, context summary, run tests + build + lint + verify acceptance criteria, send results to lead when done>"
+   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<verification scope, context summary, run tests + build + lint + verify acceptance criteria, no task claiming required in this phase, send results to lead when done>"
    ```
    Wait for message → team_results → fix blockers → team_shutdown (no team_merge needed, worktree:false)
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-onboard",
-  "version": "0.2.13",
+  "version": "0.3.1",
   "description": "Prepare any brownfield codebase for AI agent workflows using OpenCode, OpenSpec, and ensemble orchestration.",
   "keywords": [
     "opencode",

package/src/steps/init-openspec.js

@@ -38,7 +38,14 @@ const ENSEMBLE_SECTION = `6. **Implement via ensemble team**
       DO NOT call team_claim yourself, only agents claim tasks.
       DO NOT proceed to 6d until team_tasks_add succeeds.
 
-   **Step 6d.** Spawn all needed specialists, then kick them off in parallel.
+   **Step 6d.** Discover relevant skills, then spawn specialists.
+
+      Before spawning, scan \`.agents/skills/\` and read each \`SKILL.md\` description line.
+      Match skills to agents by domain:
+      - front-engineer: UI, components, framework skills (e.g. next-best-practices, browser-automation)
+      - back-engineer: API, data, service skills
+      - infra-engineer: cloud, pipeline, deployment skills
+      - quality-engineer: testing, coverage skills
 
       Each team_spawn MUST include the agent field (required, causes NOT NULL error if omitted).
 
@@ -49,8 +56,9 @@ const ENSEMBLE_SECTION = `6. **Implement via ensemble team**
       4. The 6 OpenCode tools they have available (these are OpenCode tools, NOT shell commands, call them directly as tools, never via bash):
          team_claim, team_tasks_complete, team_tasks_list, team_tasks_add, team_message, team_broadcast
       5. How to proceed: call team_claim tool with the task_id to claim a task before starting it, call team_tasks_complete tool after finishing it, repeat until all their tasks are done, then call team_message tool to notify lead with results or blockers
+      6. Which skills to load: list the skill names and paths they MUST read before implementing. Example: "Before starting, read \`.agents/skills/next-best-practices/SKILL.md\` and follow its rules for all Next.js code."
 
-      Keep spawn prompts under 500 tokens. Do not describe team internals or how ensemble works.
+      Keep spawn prompts under 600 tokens. Do not describe team internals or how ensemble works.
       Only spawn agents whose tasks are actually needed by this change. Skip agents with no tasks.
 
       First spawn all agents (wait for each team_spawn to confirm before the next):
@@ -65,9 +73,9 @@ const ENSEMBLE_SECTION = `6. **Implement via ensemble team**
 
       Then immediately send each spawned agent a start message to kick them off:
       \`\`\`
-      team_message to:"back" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"front" text:"Start now. Claim your first task with team_claim and begin implementing."
-      team_message to:"infra" text:"Start now. Claim your first task with team_claim and begin implementing."
+      team_message to:"back" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"front" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
+      team_message to:"infra" text:"Start now. Read all skills listed in your prompt first, confirm loaded skills, then claim your first task with team_claim."
       \`\`\`
 
    **Step 6e.** After sending start messages, tell the user what is running, then STOP and wait.
@@ -89,7 +97,7 @@ const ENSEMBLE_SECTION = `6. **Implement via ensemble team**
 
    Spawn quality engineer with worktree:false (read-only, no file edits):
    \`\`\`
-   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<task list, context summary, run tests + build + lint + verify acceptance criteria, send results to lead when done>"
+   team_spawn name:"quality" agent:"quality-engineer" worktree:false prompt:"<verification scope, context summary, run tests + build + lint + verify acceptance criteria, no task claiming required in this phase, send results to lead when done>"
    \`\`\`
    Wait for message -> team_results -> fix blockers -> team_shutdown (no team_merge needed, worktree:false)