[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M76VZ6ATXT37D6RF22FBTXGT":3,"skill-same:01M76VZ6ATXT37D6RF22FBTXGT":62},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":14,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":32,"updatedAt":33,"install":34,"resources":47},"01M76VZ6ATXT37D6RF22FBTXGT","ag-team",null,"Orchestrate Agent Teams for parallel multi-session collaboration. Use for research, implementation, review, and debug workflows requiring independent teammates.","ngxccc\u002Fagent-skills-kit","ngxccc","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F163379246?v=4","01M758XVG64QK26AGK0WKB26TB","Ngoc Tran","01M758XVHB8J6MHV0YTBVDBXCA",0,"MIT","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fag-team\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: ag-team\ndescription: \"Orchestrate Agent Teams for parallel multi-session collaboration. Use for research, implementation, review, and debug workflows requiring independent teammates.\"\ntrigger_keywords: team\nlayer: helper\n---\n\n# Agent Teams - Parallel Orchestration Utility\n\nCoordinate multiple independent Claude Code sessions. Each teammate has own context window, loads project context (CLAUDE.md, skills, agents), communicates via shared task list and messaging.\n\n**Requires:** Agent Teams enabled. Set `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` in settings.json env.\n**Requires:** CLI terminal — `TaskCreate`\u002F`TaskUpdate`\u002F`TaskGet`\u002F`TaskList` and `TeamCreate`\u002F`TeamDelete` are **disabled in VSCode extension** (`isTTY` check). Agent Teams CANNOT run in VSCode.\n**Model requirement:** All teammates must run Opus 4.6 (Agent Teams constraint).\n\n## When to Apply\n\nUse when orchestrating agent teams for parallel multi-session collaboration.\n\n## How to Use\n\n## Usage\n\n```\n\u002Fag-team \u003Ctemplate> \u003Ccontext> [flags]\n```\n\n**Templates:** `research`, `execute`, `review`, `debug`\n\n**Flags:**\n\n- `--devs N` | `--researchers N` | `--reviewers N` | `--debuggers N` -- team size\n- `--plan-approval` \u002F `--no-plan-approval` -- plan gate (default: on for execute)\n- `--delegate` -- lead only coordinates, never touches code\n- `--worktree` -- use git worktrees for implementation isolation (default: on for execute)\n\n## Execution Protocol\n\n**Pre-flight (MANDATORY -- merged into step 2 of every template):**\n\n1. Step 2 of every template calls `TeamCreate(team_name: \"...\", ...)`. Do NOT check whether the tool exists first -- just call it.\n2. If the call SUCCEEDS: continue with the template.\n3. If the call returns an ERROR or is unrecognized: **STOP. Tell user:** \"Agent Teams requires `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` in settings.json. Team mode is not available.\"\n4. Do NOT fall back to subagents. `\u002Fag-team` MUST use Agent Teams or abort.\n5. Ensure `TeamCreate` was called before spawning teammates -- team association happens via session context.\n\nWhen activated, first confirm team creation if the user has not already explicitly approved using agent teams for this task.\nDo not use this skill to bypass the repo's normal human approval gates.\nExecute the tool calls in order only after team usage is approved. Report progress after each major step.\n\n### --delegate Mode\n\nWhen `--delegate` flag is passed:\n\n- Lead enters delegate mode (`Shift+Tab` after TeamCreate)\n- Lead ONLY: spawns teammates, manages tasks, sends messages, synthesizes reports\n- Lead NEVER: edits files, runs tests, executes git commands directly\n- For execute Step 7 MERGE: spawn a dedicated merge teammate instead of lead doing it\n- For all templates: lead coordinates and reports, delegates ALL implementation work\n\n---\n\n## Tool Reference (Quick)\n\n### Agent Tool (spawn teammates)\n\n```\nAgent(\n  subagent_type: \"research-agent\" | \"execute-agent\" | \"code-reviewer\" | \"debugger\" | \"tester\" | ...,\n  description: \"short task summary\",\n  prompt: \"full instructions + CK Context Block\",\n  model: \"opus\",                    # Required for Agent Teams teammates\n  run_in_background: true,          # Non-blocking spawn\n  isolation: \"worktree\"             # Git worktree isolation (execute devs)\n)\n```\n\n**Note:** `Task` was renamed to `Agent` in v2.1.63. Both names work; prefer `Agent` for new code.\n\n### Team Management Tools\n\n| Tool          | Purpose                    | Params                                                            |\n| ------------- | -------------------------- | ----------------------------------------------------------------- |\n| `TeamCreate`  | Create team + task list    | `team_name`, `description`                                        |\n| `TeamDelete`  | Remove team resources      | _none_ -- just call it                                            |\n| `TaskCreate`  | Create work item           | `subject`, `description`, `priority`, `addBlockedBy`, `addBlocks` |\n| `TaskUpdate`  | Claim\u002Fcomplete task        | `taskId`, `status`, `owner`, `metadata`                           |\n| `TaskGet`     | Full task details          | `taskId`                                                          |\n| `TaskList`    | All tasks (minimal fields) | _none_                                                            |\n| `SendMessage` | Inter-agent messaging      | `type`, `to`\u002F`recipient`, `message`                               |\n\n### SendMessage Types\n\n| Type                     | Purpose                                                     |\n| ------------------------ | ----------------------------------------------------------- |\n| `message`                | DM to one teammate (requires `recipient`)                   |\n| `broadcast`              | Send to ALL teammates (use sparingly)                       |\n| `shutdown_request`       | Ask teammate to gracefully exit                             |\n| `shutdown_response`      | Teammate approves\u002Frejects shutdown (requires `request_id`)  |\n| `plan_approval_response` | Lead approves\u002Frejects teammate plan (requires `request_id`) |\n\n---\n\n## CK Context Block\n\nEvery teammate spawn prompt MUST include this context at the end:\n\n```\nCK Context:\n- Work dir: {CK_PROJECT_ROOT or CWD}\n- Feature: {CK_FEATURE or \"none\"}\n- Reports: {CK_REPORTS_PATH or \"process\u002Fgeneral-plans\u002Freports\u002F\"}\n- Plans: {CK_PLANS_PATH or \"process\u002Fgeneral-plans\u002Factive\u002F\"}\n- Branch: {CK_GIT_BRANCH or current branch}\n- Naming: {CK_NAME_PATTERN or \"YYMMDD-HHMM\"}\n- Active plan: {CK_ACTIVE_PLAN or \"none\"}\n- Commits: conventional (feat:, fix:, docs:, refactor:, test:, chore:)\n- Refer to teammates by NAME, not agent ID\n```\n\n---\n\n## ON `\u002Fag-team research \u003Ctopic>` [--researchers N]:\n\n_Coordinates parallel `research-agent` teammates plus shared helper skills where useful._\n\nIMMEDIATELY execute in order:\n\n1. **Derive N angles** from `\u003Ctopic>` (default N=3):\n   - Angle 1: Architecture, patterns, proven approaches\n   - Angle 2: Alternatives, competing solutions, trade-offs\n   - Angle 3: Risks, edge cases, failure modes, security\n   - (If N>3, derive additional angles from topic context)\n\n2. **CALL** `TeamCreate(team_name: \"\u003Ctopic-slug>\")` after user approval to use agent teams\n\n3. **CALL** `TaskCreate` x N -- one per angle:\n   - Subject: `Research: \u003Cangle-title>`\n   - Description: `Investigate \u003Cangle> for topic: \u003Ctopic>. Save report to: {CK_REPORTS_PATH}\u002Fresearch-agent-{N}-{CK_NAME_PATTERN}-{topic-slug}.md. Format: Executive summary, key findings, evidence, contradictions, unresolved questions. Do not recommend implementations. Mark task completed when done. Send findings summary to lead.`\n\n4. **SPAWN** teammates x N via `Agent` tool:\n   - `subagent_type: \"research-agent\"`, `model: \"opus\"`\n   - `run_in_background: true` (non-blocking -- spawn all N concurrently)\n   - `name: \"research-agent-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted hook events + TaskList fallback:\n   - TaskCompleted events auto-notify when `research-agent` subagents finish\n   - Fallback: Check TaskList if no event received in 60s\n   - If stuck >5 min, message teammate directly\n\n6. **READ** all `research-agent` reports from `{CK_REPORTS_PATH}\u002F`\n\n7. **SYNTHESIZE** into: `{CK_REPORTS_PATH}\u002Fresearch-summary-{CK_NAME_PATTERN}-{topic-slug}.md`\n   Format: exec summary, key findings, comparative analysis, trade-off evidence, unresolved questions, decision inputs for the orchestrator.\n\n8. **SHUTDOWN**: `SendMessage(type: \"shutdown_request\")` to each teammate\n\n9. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n10. **REPORT**: Tell user `Research complete. Summary: {path}. N reports generated.`\n11. **FOLLOW-UP**: If durable process\u002Fcontext learnings surfaced, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team execute \u003Cplan-path>` [--devs N]:\n\n_Coordinates approved-plan execution with parallel `execute-agent` teammates, testing, review, and final reporting without introducing a separate execution owner._\n\nDo not use this template to flatten an entire multi-phase program into one execution wave. If the\nselected work is really a phase program, the lead must first pick one current phase plan and execute\nthat phase only. See `process\u002Fdevelopment-protocols\u002Fphase-programs.md`.\n\nIMMEDIATELY execute in order:\n\n1. **PRE-FLIGHT APPROVAL CHECK**:\n   - Require one explicit approved plan file path under `process\u002Fgeneral-plans\u002Factive\u002F` or `process\u002Ffeatures\u002F*\u002Factive\u002F`\n   - If the user supplied only a description or an unapproved draft, STOP and tell the user to create\u002Fselect a plan first, then return with `ENTER EXECUTE MODE`\n   - Do not use `ag-team execute` to bypass the repo's explicit execute approval gate\n   - If the selected plan is an umbrella\u002Forchestration plan for a large program, STOP and require\n     one explicit current phase plan instead of executing the whole program at once\n\n2. **READ** the selected plan and parse it into N independent task groups with file ownership boundaries\n\n3. **CALL** `TeamCreate(team_name: \"\u003Cfeature-slug>\")` after user approval to use agent teams\n\n4. **CALL** `TaskCreate` x (N + 1) -- N dev tasks + 1 tester task:\n   - Dev tasks: include `File ownership: \u003Cglob patterns>` -- NO overlap between devs\n   - Tester task: `addBlockedBy` all dev task IDs\n   - Each task description includes: implementation scope, file ownership, acceptance criteria\n\n5. **SPAWN** developer teammates x N via `Agent` tool:\n   - `subagent_type: \"execute-agent\"`, `model: \"opus\"`\n   - `isolation: \"worktree\"` -- each dev gets isolated git worktree (no file conflicts)\n   - `run_in_background: true`\n   - `name: \"dev-{N}\"`\n   - Prompt: task description + exact plan path + CK Context Block\n   - If `--plan-approval`: require a short task-level implementation outline before coding, but do not reopen the top-level execute approval gate\n   - REVIEW and APPROVE each developer's task outline via `plan_approval_response` before code changes\n\n6. **MONITOR** dev completion via TaskCompleted events:\n   - TaskCompleted hook notifies when each dev task finishes\n   - When all N dev tasks show completed, spawn tester immediately\n   - TeammateIdle events confirm devs are available for shutdown\n   - Fallback: Check TaskList if no events received in 60s\n   - Spawn tester: `Agent(subagent_type: \"tester\", model: \"opus\", name: \"tester\")`\n   - Tester runs full test suite, reports pass\u002Ffail\n\n7. **MERGE** worktree branches (if `isolation: \"worktree\"` was used):\n   - Discover branches: check Agent result for branch names, or `git worktree list`\n   - In `--delegate` mode, spawn a dedicated merge teammate for this step\n   - Otherwise, for each dev branch: `git merge \u003Cdev-branch> --no-ff`\n   - If conflict: resolve manually (lead owns shared files), then `git add . && git merge --continue`\n   - Cleanup: `git worktree remove \u003Cpath>` for each worktree\n   - Verify: `git log --oneline --graph` to confirm merge topology\n\n8. **DOCS SYNC EVAL**:\n\n   ```\n   Docs impact: [none|minor|major]\n   Action: [no update needed -- \u003Creason>] | [updated \u003Cpage>] | [needs separate PR]\n   ```\n\n9. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n10. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n11. **REPORT**: Tell user what was implemented, test results, docs impact, and any remaining concerns.\n12. **FOLLOW-UP**: If durable process\u002Fcontext learnings surfaced, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team review \u003Cscope>` [--reviewers N]:\n\n_Coordinates parallel `code-reviewer` teammates using the absorbed review methodology now owned by the agent path._\n\nIMMEDIATELY execute in order:\n\n1. **DERIVE** N review focuses from `\u003Cscope>` (default N=3):\n   - Focus 1: Security -- vulnerabilities, auth, input validation, OWASP\n   - Focus 2: Performance -- bottlenecks, memory, complexity, scaling\n   - Focus 3: Test coverage -- gaps, edge cases, error paths\n   - (If N>3, derive from scope: architecture, DX, accessibility, etc.)\n\n2. **CALL** `TeamCreate(team_name: \"review-\u003Cscope-slug>\")` after user approval to use agent teams\n\n3. **CALL** `TaskCreate` x N -- one per focus:\n   - Subject: `Review: \u003Cfocus-title>`\n   - Description: `Review \u003Cscope> for \u003Cfocus>. Output severity-rated findings only. Format: [CRITICAL|IMPORTANT|MODERATE] \u003Cfinding> -- \u003Cevidence> -- \u003Crecommendation>. No \"seems\" or \"probably\" -- concrete evidence only. Save to: {CK_REPORTS_PATH}\u002Freviewer-{N}-{CK_NAME_PATTERN}-{scope-slug}.md. Mark task completed when done.`\n\n4. **SPAWN** reviewers x N via `Agent` tool:\n   - `subagent_type: \"code-reviewer\"`, `model: \"opus\"`\n   - `run_in_background: true`\n   - `name: \"reviewer-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted hook events + TaskList fallback:\n   - TaskCompleted events auto-notify when reviewers finish\n   - Fallback: Check TaskList if no event received in 60s\n\n6. **SYNTHESIZE** into: `{CK_REPORTS_PATH}\u002Freview-{scope-slug}.md`\n   - Deduplicate findings across reviewers\n   - Prioritize by severity: CRITICAL > IMPORTANT > MODERATE\n   - Create action items list with owners\n\n7. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n8. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n9. **REPORT**: Tell user `Review complete. {X} findings ({Y} critical). Report: {path}.`\n10. **FOLLOW-UP**: If the review changes durable process\u002Fcontext guidance, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team debug \u003Cissue>` [--debuggers N]:\n\n_Coordinates parallel `debugger` teammates using the absorbed root-cause-first debug workflow now owned by the debugger path._\n\nIMMEDIATELY execute in order:\n\n1. **GENERATE** N competing hypotheses from `\u003Cissue>` (default N=3):\n   - Each hypothesis must be independently testable\n   - Each must predict different observable symptoms\n   - Frame as: \"If \u003Ccause>, then we should see \u003Cevidence>\"\n\n2. **CALL** `TeamCreate(team_name: \"debug-\u003Cissue-slug>\")`\n\n3. **CALL** `TaskCreate` x N -- one per hypothesis:\n   - Subject: `Debug: Test hypothesis -- \u003Ctheory>`\n   - Description: `Investigate hypothesis: \u003Ctheory>. For issue: \u003Cissue>. ADVERSARIAL: actively try to disprove other theories. Message other debuggers to challenge findings. Report evidence FOR and AGAINST your theory. Save findings to: {CK_REPORTS_PATH}\u002Fdebugger-{N}-{CK_NAME_PATTERN}-{issue-slug}.md. Mark task completed when done.`\n\n4. **SPAWN** debugger teammates x N via `Agent` tool:\n   - `subagent_type: \"debugger\"`, `model: \"opus\"`\n   - `run_in_background: true`\n   - `name: \"debugger-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted events. Debuggers should message each other -- let them converge.\n   - TaskCompleted events notify as each hypothesis is tested\n   - TeammateIdle events indicate debugger awaiting peer input\n   - Fallback: Check TaskList if no events in 60s\n\n6. **READ** all debugger reports. Identify surviving theory as root cause.\n\n7. **WRITE** root cause report: `{CK_REPORTS_PATH}\u002Fdebug-{issue-slug}.md`\n   Format: Root cause, evidence chain, disproven hypotheses, recommended fix.\n\n8. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n9. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n10. **REPORT**: Tell user `Debug complete. Root cause: \u003Csummary>. Report: {path}.`\n11. **FOLLOW-UP**: If the debug session produced durable process\u002Fcontext learnings, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## When to Apply Agent Teams vs Subagents\n\n| Scenario | Subagents (Agent tool) | Agent Teams |\n| -------- | ---------------------- | ----------- |\n\n## How to Use\n\nRefer to the workflow instructions and command references detailed below.\n| Focused task (test, lint, single review) | **Yes** | Overkill |\n| Sequential chain (plan -> code -> test) | **Yes** | No |\n| 3+ independent parallel workstreams | Maybe | **Yes** |\n| Competing debug hypotheses | No | **Yes** |\n| Cross-layer work (FE + BE + tests) | Maybe | **Yes** |\n| Workers need to discuss\u002Fchallenge findings | No | **Yes** |\n| Token budget is tight | **Yes** | No (high cost) |\n\n## Token Budget\n\n| Template     | Estimated Tokens | Notes                           |\n| ------------ | ---------------- | ------------------------------- |\n| Research (3) | ~150K-300K       | Read-only, moderate cost        |\n| Execute (4)  | ~400K-800K       | Highest cost -- code generation |\n| Review (3)   | ~100K-200K       | Read-only, moderate cost        |\n| Debug (3)    | ~200K-400K       | Mixed read\u002Fexecute              |\n\n## Agent Memory\n\nTeammates with `memory: project` in their agent definition retain learnings across team sessions. Memory persists in `.claude\u002Fagent-memory\u002F\u003Cname>\u002F` (gitignored). Useful for:\n\n- Code reviewer remembering project conventions\n- Debugger recalling past failure patterns\n- Tester tracking flaky tests and coverage gaps\n- Researcher accumulating domain knowledge\n\nMemory persists after team cleanup -- it's in `.claude\u002Fagent-memory\u002F`, not `~\u002F.claude\u002Fteams\u002F`.\n\n## Worktree Isolation (Execute Template)\n\nFor implementation teams, `isolation: \"worktree\"` on the Agent tool gives each dev:\n\n- **Own git worktree** -- isolated working directory, staging area, HEAD\n- **Own branch** -- auto-created, returned in agent result\n- **No file conflicts** -- devs can edit same files independently\n- **Safe parallel editing** -- `.git` dir shared, everything else isolated\n\nAfter all devs complete, lead merges branches sequentially. This is the safest pattern for parallel code changes.\n\n## Error Recovery\n\n1. **Check status**: `Shift+Up\u002FDown` (in-process) or click pane (split)\n2. **Redirect**: Send direct message with corrective instructions\n3. **Replace**: Shut down failed teammate, spawn replacement for same task\n4. **Reassign**: `TaskUpdate` stuck task to unblock dependents\n\n## Abort Team\n\n```\nShut down all teammates. Then call TeamDelete (no parameters).\n```\n\nIf unresponsive: close terminal or kill session. Clean orphaned configs at `~\u002F.claude\u002Fteams\u002F` manually.\n\n## Display Modes\n\n- **auto** (default): split panes if in tmux, otherwise in-process\n- **in-process**: all in one terminal. `Shift+Up\u002FDown` navigate. `Ctrl+T` task list.\n- **tmux\u002Fsplit**: each teammate own pane. Requires tmux or iTerm2.\n\n## Rules Reference\n\nSee `process\u002Fdevelopment-protocols\u002Forchestration.md` for teammate coordination, status handling, and context-isolation rules.\n\n> v3.0.0: Agent tool migration, worktree isolation for execute devs, run_in_background spawning, updated model requirements.\n\n## References\n\n- [process\u002Fcontext\u002Fall-context.md](process\u002Fcontext\u002Fall-context.md)\n","# Agent Teams - Parallel Orchestration Utility\n\nCoordinate multiple independent Claude Code sessions. Each teammate has own context window, loads project context (CLAUDE.md, skills, agents), communicates via shared task list and messaging.\n\n**Requires:** Agent Teams enabled. Set `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` in settings.json env.\n**Requires:** CLI terminal — `TaskCreate`\u002F`TaskUpdate`\u002F`TaskGet`\u002F`TaskList` and `TeamCreate`\u002F`TeamDelete` are **disabled in VSCode extension** (`isTTY` check). Agent Teams CANNOT run in VSCode.\n**Model requirement:** All teammates must run Opus 4.6 (Agent Teams constraint).\n\n## When to Apply\n\nUse when orchestrating agent teams for parallel multi-session collaboration.\n\n## How to Use\n\n## Usage\n\n```\n\u002Fag-team \u003Ctemplate> \u003Ccontext> [flags]\n```\n\n**Templates:** `research`, `execute`, `review`, `debug`\n\n**Flags:**\n\n- `--devs N` | `--researchers N` | `--reviewers N` | `--debuggers N` -- team size\n- `--plan-approval` \u002F `--no-plan-approval` -- plan gate (default: on for execute)\n- `--delegate` -- lead only coordinates, never touches code\n- `--worktree` -- use git worktrees for implementation isolation (default: on for execute)\n\n## Execution Protocol\n\n**Pre-flight (MANDATORY -- merged into step 2 of every template):**\n\n1. Step 2 of every template calls `TeamCreate(team_name: \"...\", ...)`. Do NOT check whether the tool exists first -- just call it.\n2. If the call SUCCEEDS: continue with the template.\n3. If the call returns an ERROR or is unrecognized: **STOP. Tell user:** \"Agent Teams requires `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` in settings.json. Team mode is not available.\"\n4. Do NOT fall back to subagents. `\u002Fag-team` MUST use Agent Teams or abort.\n5. Ensure `TeamCreate` was called before spawning teammates -- team association happens via session context.\n\nWhen activated, first confirm team creation if the user has not already explicitly approved using agent teams for this task.\nDo not use this skill to bypass the repo's normal human approval gates.\nExecute the tool calls in order only after team usage is approved. Report progress after each major step.\n\n### --delegate Mode\n\nWhen `--delegate` flag is passed:\n\n- Lead enters delegate mode (`Shift+Tab` after TeamCreate)\n- Lead ONLY: spawns teammates, manages tasks, sends messages, synthesizes reports\n- Lead NEVER: edits files, runs tests, executes git commands directly\n- For execute Step 7 MERGE: spawn a dedicated merge teammate instead of lead doing it\n- For all templates: lead coordinates and reports, delegates ALL implementation work\n\n---\n\n## Tool Reference (Quick)\n\n### Agent Tool (spawn teammates)\n\n```\nAgent(\n  subagent_type: \"research-agent\" | \"execute-agent\" | \"code-reviewer\" | \"debugger\" | \"tester\" | ...,\n  description: \"short task summary\",\n  prompt: \"full instructions + CK Context Block\",\n  model: \"opus\",                    # Required for Agent Teams teammates\n  run_in_background: true,          # Non-blocking spawn\n  isolation: \"worktree\"             # Git worktree isolation (execute devs)\n)\n```\n\n**Note:** `Task` was renamed to `Agent` in v2.1.63. Both names work; prefer `Agent` for new code.\n\n### Team Management Tools\n\n| Tool          | Purpose                    | Params                                                            |\n| ------------- | -------------------------- | ----------------------------------------------------------------- |\n| `TeamCreate`  | Create team + task list    | `team_name`, `description`                                        |\n| `TeamDelete`  | Remove team resources      | _none_ -- just call it                                            |\n| `TaskCreate`  | Create work item           | `subject`, `description`, `priority`, `addBlockedBy`, `addBlocks` |\n| `TaskUpdate`  | Claim\u002Fcomplete task        | `taskId`, `status`, `owner`, `metadata`                           |\n| `TaskGet`     | Full task details          | `taskId`                                                          |\n| `TaskList`    | All tasks (minimal fields) | _none_                                                            |\n| `SendMessage` | Inter-agent messaging      | `type`, `to`\u002F`recipient`, `message`                               |\n\n### SendMessage Types\n\n| Type                     | Purpose                                                     |\n| ------------------------ | ----------------------------------------------------------- |\n| `message`                | DM to one teammate (requires `recipient`)                   |\n| `broadcast`              | Send to ALL teammates (use sparingly)                       |\n| `shutdown_request`       | Ask teammate to gracefully exit                             |\n| `shutdown_response`      | Teammate approves\u002Frejects shutdown (requires `request_id`)  |\n| `plan_approval_response` | Lead approves\u002Frejects teammate plan (requires `request_id`) |\n\n---\n\n## CK Context Block\n\nEvery teammate spawn prompt MUST include this context at the end:\n\n```\nCK Context:\n- Work dir: {CK_PROJECT_ROOT or CWD}\n- Feature: {CK_FEATURE or \"none\"}\n- Reports: {CK_REPORTS_PATH or \"process\u002Fgeneral-plans\u002Freports\u002F\"}\n- Plans: {CK_PLANS_PATH or \"process\u002Fgeneral-plans\u002Factive\u002F\"}\n- Branch: {CK_GIT_BRANCH or current branch}\n- Naming: {CK_NAME_PATTERN or \"YYMMDD-HHMM\"}\n- Active plan: {CK_ACTIVE_PLAN or \"none\"}\n- Commits: conventional (feat:, fix:, docs:, refactor:, test:, chore:)\n- Refer to teammates by NAME, not agent ID\n```\n\n---\n\n## ON `\u002Fag-team research \u003Ctopic>` [--researchers N]:\n\n_Coordinates parallel `research-agent` teammates plus shared helper skills where useful._\n\nIMMEDIATELY execute in order:\n\n1. **Derive N angles** from `\u003Ctopic>` (default N=3):\n   - Angle 1: Architecture, patterns, proven approaches\n   - Angle 2: Alternatives, competing solutions, trade-offs\n   - Angle 3: Risks, edge cases, failure modes, security\n   - (If N>3, derive additional angles from topic context)\n\n2. **CALL** `TeamCreate(team_name: \"\u003Ctopic-slug>\")` after user approval to use agent teams\n\n3. **CALL** `TaskCreate` x N -- one per angle:\n   - Subject: `Research: \u003Cangle-title>`\n   - Description: `Investigate \u003Cangle> for topic: \u003Ctopic>. Save report to: {CK_REPORTS_PATH}\u002Fresearch-agent-{N}-{CK_NAME_PATTERN}-{topic-slug}.md. Format: Executive summary, key findings, evidence, contradictions, unresolved questions. Do not recommend implementations. Mark task completed when done. Send findings summary to lead.`\n\n4. **SPAWN** teammates x N via `Agent` tool:\n   - `subagent_type: \"research-agent\"`, `model: \"opus\"`\n   - `run_in_background: true` (non-blocking -- spawn all N concurrently)\n   - `name: \"research-agent-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted hook events + TaskList fallback:\n   - TaskCompleted events auto-notify when `research-agent` subagents finish\n   - Fallback: Check TaskList if no event received in 60s\n   - If stuck >5 min, message teammate directly\n\n6. **READ** all `research-agent` reports from `{CK_REPORTS_PATH}\u002F`\n\n7. **SYNTHESIZE** into: `{CK_REPORTS_PATH}\u002Fresearch-summary-{CK_NAME_PATTERN}-{topic-slug}.md`\n   Format: exec summary, key findings, comparative analysis, trade-off evidence, unresolved questions, decision inputs for the orchestrator.\n\n8. **SHUTDOWN**: `SendMessage(type: \"shutdown_request\")` to each teammate\n\n9. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n10. **REPORT**: Tell user `Research complete. Summary: {path}. N reports generated.`\n11. **FOLLOW-UP**: If durable process\u002Fcontext learnings surfaced, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team execute \u003Cplan-path>` [--devs N]:\n\n_Coordinates approved-plan execution with parallel `execute-agent` teammates, testing, review, and final reporting without introducing a separate execution owner._\n\nDo not use this template to flatten an entire multi-phase program into one execution wave. If the\nselected work is really a phase program, the lead must first pick one current phase plan and execute\nthat phase only. See `process\u002Fdevelopment-protocols\u002Fphase-programs.md`.\n\nIMMEDIATELY execute in order:\n\n1. **PRE-FLIGHT APPROVAL CHECK**:\n   - Require one explicit approved plan file path under `process\u002Fgeneral-plans\u002Factive\u002F` or `process\u002Ffeatures\u002F*\u002Factive\u002F`\n   - If the user supplied only a description or an unapproved draft, STOP and tell the user to create\u002Fselect a plan first, then return with `ENTER EXECUTE MODE`\n   - Do not use `ag-team execute` to bypass the repo's explicit execute approval gate\n   - If the selected plan is an umbrella\u002Forchestration plan for a large program, STOP and require\n     one explicit current phase plan instead of executing the whole program at once\n\n2. **READ** the selected plan and parse it into N independent task groups with file ownership boundaries\n\n3. **CALL** `TeamCreate(team_name: \"\u003Cfeature-slug>\")` after user approval to use agent teams\n\n4. **CALL** `TaskCreate` x (N + 1) -- N dev tasks + 1 tester task:\n   - Dev tasks: include `File ownership: \u003Cglob patterns>` -- NO overlap between devs\n   - Tester task: `addBlockedBy` all dev task IDs\n   - Each task description includes: implementation scope, file ownership, acceptance criteria\n\n5. **SPAWN** developer teammates x N via `Agent` tool:\n   - `subagent_type: \"execute-agent\"`, `model: \"opus\"`\n   - `isolation: \"worktree\"` -- each dev gets isolated git worktree (no file conflicts)\n   - `run_in_background: true`\n   - `name: \"dev-{N}\"`\n   - Prompt: task description + exact plan path + CK Context Block\n   - If `--plan-approval`: require a short task-level implementation outline before coding, but do not reopen the top-level execute approval gate\n   - REVIEW and APPROVE each developer's task outline via `plan_approval_response` before code changes\n\n6. **MONITOR** dev completion via TaskCompleted events:\n   - TaskCompleted hook notifies when each dev task finishes\n   - When all N dev tasks show completed, spawn tester immediately\n   - TeammateIdle events confirm devs are available for shutdown\n   - Fallback: Check TaskList if no events received in 60s\n   - Spawn tester: `Agent(subagent_type: \"tester\", model: \"opus\", name: \"tester\")`\n   - Tester runs full test suite, reports pass\u002Ffail\n\n7. **MERGE** worktree branches (if `isolation: \"worktree\"` was used):\n   - Discover branches: check Agent result for branch names, or `git worktree list`\n   - In `--delegate` mode, spawn a dedicated merge teammate for this step\n   - Otherwise, for each dev branch: `git merge \u003Cdev-branch> --no-ff`\n   - If conflict: resolve manually (lead owns shared files), then `git add . && git merge --continue`\n   - Cleanup: `git worktree remove \u003Cpath>` for each worktree\n   - Verify: `git log --oneline --graph` to confirm merge topology\n\n8. **DOCS SYNC EVAL**:\n\n   ```\n   Docs impact: [none|minor|major]\n   Action: [no update needed -- \u003Creason>] | [updated \u003Cpage>] | [needs separate PR]\n   ```\n\n9. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n10. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n11. **REPORT**: Tell user what was implemented, test results, docs impact, and any remaining concerns.\n12. **FOLLOW-UP**: If durable process\u002Fcontext learnings surfaced, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team review \u003Cscope>` [--reviewers N]:\n\n_Coordinates parallel `code-reviewer` teammates using the absorbed review methodology now owned by the agent path._\n\nIMMEDIATELY execute in order:\n\n1. **DERIVE** N review focuses from `\u003Cscope>` (default N=3):\n   - Focus 1: Security -- vulnerabilities, auth, input validation, OWASP\n   - Focus 2: Performance -- bottlenecks, memory, complexity, scaling\n   - Focus 3: Test coverage -- gaps, edge cases, error paths\n   - (If N>3, derive from scope: architecture, DX, accessibility, etc.)\n\n2. **CALL** `TeamCreate(team_name: \"review-\u003Cscope-slug>\")` after user approval to use agent teams\n\n3. **CALL** `TaskCreate` x N -- one per focus:\n   - Subject: `Review: \u003Cfocus-title>`\n   - Description: `Review \u003Cscope> for \u003Cfocus>. Output severity-rated findings only. Format: [CRITICAL|IMPORTANT|MODERATE] \u003Cfinding> -- \u003Cevidence> -- \u003Crecommendation>. No \"seems\" or \"probably\" -- concrete evidence only. Save to: {CK_REPORTS_PATH}\u002Freviewer-{N}-{CK_NAME_PATTERN}-{scope-slug}.md. Mark task completed when done.`\n\n4. **SPAWN** reviewers x N via `Agent` tool:\n   - `subagent_type: \"code-reviewer\"`, `model: \"opus\"`\n   - `run_in_background: true`\n   - `name: \"reviewer-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted hook events + TaskList fallback:\n   - TaskCompleted events auto-notify when reviewers finish\n   - Fallback: Check TaskList if no event received in 60s\n\n6. **SYNTHESIZE** into: `{CK_REPORTS_PATH}\u002Freview-{scope-slug}.md`\n   - Deduplicate findings across reviewers\n   - Prioritize by severity: CRITICAL > IMPORTANT > MODERATE\n   - Create action items list with owners\n\n7. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n8. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n9. **REPORT**: Tell user `Review complete. {X} findings ({Y} critical). Report: {path}.`\n10. **FOLLOW-UP**: If the review changes durable process\u002Fcontext guidance, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## ON `\u002Fag-team debug \u003Cissue>` [--debuggers N]:\n\n_Coordinates parallel `debugger` teammates using the absorbed root-cause-first debug workflow now owned by the debugger path._\n\nIMMEDIATELY execute in order:\n\n1. **GENERATE** N competing hypotheses from `\u003Cissue>` (default N=3):\n   - Each hypothesis must be independently testable\n   - Each must predict different observable symptoms\n   - Frame as: \"If \u003Ccause>, then we should see \u003Cevidence>\"\n\n2. **CALL** `TeamCreate(team_name: \"debug-\u003Cissue-slug>\")`\n\n3. **CALL** `TaskCreate` x N -- one per hypothesis:\n   - Subject: `Debug: Test hypothesis -- \u003Ctheory>`\n   - Description: `Investigate hypothesis: \u003Ctheory>. For issue: \u003Cissue>. ADVERSARIAL: actively try to disprove other theories. Message other debuggers to challenge findings. Report evidence FOR and AGAINST your theory. Save findings to: {CK_REPORTS_PATH}\u002Fdebugger-{N}-{CK_NAME_PATTERN}-{issue-slug}.md. Mark task completed when done.`\n\n4. **SPAWN** debugger teammates x N via `Agent` tool:\n   - `subagent_type: \"debugger\"`, `model: \"opus\"`\n   - `run_in_background: true`\n   - `name: \"debugger-{N}\"`\n   - Prompt: task description + CK Context Block\n\n5. **MONITOR** via TaskCompleted events. Debuggers should message each other -- let them converge.\n   - TaskCompleted events notify as each hypothesis is tested\n   - TeammateIdle events indicate debugger awaiting peer input\n   - Fallback: Check TaskList if no events in 60s\n\n6. **READ** all debugger reports. Identify surviving theory as root cause.\n\n7. **WRITE** root cause report: `{CK_REPORTS_PATH}\u002Fdebug-{issue-slug}.md`\n   Format: Root cause, evidence chain, disproven hypotheses, recommended fix.\n\n8. **SHUTDOWN** all teammates via `SendMessage(type: \"shutdown_request\")`\n9. **CLEANUP**: `TeamDelete` (no parameters -- just call it)\n\n10. **REPORT**: Tell user `Debug complete. Root cause: \u003Csummary>. Report: {path}.`\n11. **FOLLOW-UP**: If the debug session produced durable process\u002Fcontext learnings, suggest `ENTER UPDATE PROCESS MODE`.\n\n---\n\n## When to Apply Agent Teams vs Subagents\n\n| Scenario | Subagents (Agent tool) | Agent Teams |\n| -------- | ---------------------- | ----------- |\n\n## How to Use\n\nRefer to the workflow instructions and command references detailed below.\n| Focused task (test, lint, single review) | **Yes** | Overkill |\n| Sequential chain (plan -> code -> test) | **Yes** | No |\n| 3+ independent parallel workstreams | Maybe | **Yes** |\n| Competing debug hypotheses | No | **Yes** |\n| Cross-layer work (FE + BE + tests) | Maybe | **Yes** |\n| Workers need to discuss\u002Fchallenge findings | No | **Yes** |\n| Token budget is tight | **Yes** | No (high cost) |\n\n## Token Budget\n\n| Template     | Estimated Tokens | Notes                           |\n| ------------ | ---------------- | ------------------------------- |\n| Research (3) | ~150K-300K       | Read-only, moderate cost        |\n| Execute (4)  | ~400K-800K       | Highest cost -- code generation |\n| Review (3)   | ~100K-200K       | Read-only, moderate cost        |\n| Debug (3)    | ~200K-400K       | Mixed read\u002Fexecute              |\n\n## Agent Memory\n\nTeammates with `memory: project` in their agent definition retain learnings across team sessions. Memory persists in `.claude\u002Fagent-memory\u002F\u003Cname>\u002F` (gitignored). Useful for:\n\n- Code reviewer remembering project conventions\n- Debugger recalling past failure patterns\n- Tester tracking flaky tests and coverage gaps\n- Researcher accumulating domain knowledge\n\nMemory persists after team cleanup -- it's in `.claude\u002Fagent-memory\u002F`, not `~\u002F.claude\u002Fteams\u002F`.\n\n## Worktree Isolation (Execute Template)\n\nFor implementation teams, `isolation: \"worktree\"` on the Agent tool gives each dev:\n\n- **Own git worktree** -- isolated working directory, staging area, HEAD\n- **Own branch** -- auto-created, returned in agent result\n- **No file conflicts** -- devs can edit same files independently\n- **Safe parallel editing** -- `.git` dir shared, everything else isolated\n\nAfter all devs complete, lead merges branches sequentially. This is the safest pattern for parallel code changes.\n\n## Error Recovery\n\n1. **Check status**: `Shift+Up\u002FDown` (in-process) or click pane (split)\n2. **Redirect**: Send direct message with corrective instructions\n3. **Replace**: Shut down failed teammate, spawn replacement for same task\n4. **Reassign**: `TaskUpdate` stuck task to unblock dependents\n\n## Abort Team\n\n```\nShut down all teammates. Then call TeamDelete (no parameters).\n```\n\nIf unresponsive: close terminal or kill session. Clean orphaned configs at `~\u002F.claude\u002Fteams\u002F` manually.\n\n## Display Modes\n\n- **auto** (default): split panes if in tmux, otherwise in-process\n- **in-process**: all in one terminal. `Shift+Up\u002FDown` navigate. `Ctrl+T` task list.\n- **tmux\u002Fsplit**: each teammate own pane. Requires tmux or iTerm2.\n\n## Rules Reference\n\nSee `process\u002Fdevelopment-protocols\u002Forchestration.md` for teammate coordination, status handling, and context-isolation rules.\n\n> v3.0.0: Agent tool migration, worktree isolation for execute devs, run_in_background spawning, updated model requirements.\n\n## References\n\n- [process\u002Fcontext\u002Fall-context.md](process\u002Fcontext\u002Fall-context.md)\n",{"name":5,"layer":24,"description":7,"trigger_keywords":25},"helper","team","main","1ef0dcd00fe806a8dc4e0b7173b9db967c3ab274",".claude\u002Fskills\u002Fag-team","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit","https:\u002F\u002Fgithub.com\u002Fngxccc","https:\u002F\u002Fngxccc.github.io","Make it work, make it right, make it fast.","2026-08-02 20:59:07",{"source":8,"skillName":5,"commands":35,"prompt":45,"zipAvailable":46},[36,39,42],{"tool":37,"command":38},"npx","npx skills add ngxccc\u002Fagent-skills-kit --skill ag-team",{"tool":40,"command":41},"bunx","bunx skills add ngxccc\u002Fagent-skills-kit --skill ag-team",{"tool":43,"command":44},"pnpm","pnpm dlx skills add ngxccc\u002Fagent-skills-kit --skill ag-team","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fag-team\u002FSKILL.md",true,[48,54,58],{"path":49,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":52,"sourceHtmlUrl":53},"references\u002Fagent-teams-controls-and-modes.md","REFERENCE","参考文档",4963,"https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fag-team\u002Freferences\u002Fagent-teams-controls-and-modes.md",{"path":55,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":56,"sourceHtmlUrl":57},"references\u002Fagent-teams-examples-and-best-practices.md",8272,"https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fag-team\u002Freferences\u002Fagent-teams-examples-and-best-practices.md",{"path":59,"resourceType":50,"resourceTypeDesc":51,"sizeBytes":60,"sourceHtmlUrl":61},"references\u002Fagent-teams-official-docs.md",10495,"https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fag-team\u002Freferences\u002Fagent-teams-official-docs.md",[63,70,77,84,91,98],{"skillNo":64,"name":65,"nameZh":6,"description":66,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":67,"categories":68,"keywords":69,"publishedAt":19},"01M76W3C8X6Q5DS0364KQT7W7A","zustand","Rules and standards for state management using Zustand in React and Next.js. Use when implementing client-side stores, persistence with localStorage, or preventing SSR hydration mismatch.","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Fzustand\u002FSKILL.md",[],[],{"skillNo":71,"name":72,"nameZh":6,"description":73,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":19},"01M76W39FJGES309VAH7SFWA9Y","zod","Zod v4 schema validation best practices. Use when defining schemas, utilizing safeParse, or customizing error messages.","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Fzod\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":19},"01M76W36HTYA4BR5FKCWMX1RFR","upstash-ratelimit","Rules, standards, and references for rate limiting in serverless environments using Upstash Ratelimit. Use when implementing rate limiting, configuring Upstash Redis, or handling request blocking.","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Fupstash-ratelimit\u002FSKILL.md",[],[],{"skillNo":85,"name":86,"nameZh":6,"description":87,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":88,"categories":89,"keywords":90,"publishedAt":19},"01M76W32H4PS01WFRS7FNPK2G1","ui-design","UI\u002FUX and frontend design best practices guidelines (formerly frontend-design). This skill should be used when writing, reviewing, or designing frontend code to ensure accessibility, performance, and usability. Triggers on tasks involving HTML structure, CSS styling, responsive layouts, form design, animations, or accessibility improvements.","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Fui-design\u002FSKILL.md",[],[],{"skillNo":92,"name":93,"nameZh":6,"description":94,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":95,"categories":96,"keywords":97,"publishedAt":19},"01M76W2YAP6RK8N0EVWFG54VD7","typescript","TypeScript performance, tsconfig, type errors, async patterns — triggered when the user asks to \"optimize TypeScript performance\", \"speed up tsc compilation\", \"configure tsconfig.json\", \"fix type errors\", \"improve async patterns\", or encounters TS errors (TS2322, TS2339, \"is not assignable to\"). Also triggers on .ts, .tsx, .d.ts file work involving type definitions, module organization, or memory management. Does NOT cover TypeScript basics, framework-specific patterns, or testing.","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Ftypescript\u002FSKILL.md",[],[],{"skillNo":99,"name":100,"nameZh":6,"description":101,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":102,"categories":103,"keywords":104,"publishedAt":19},"01M76W2RMAZNNJQ2HECH3X6ZSH","typescript-refactor","TypeScript and TSX refactoring and modernization guidelines from a principal specialist perspective, current to TypeScript 6.0 and React 19. This skill should be used when refactoring, reviewing, or modernizing TypeScript or React\u002FTSX code for type safety, compiler performance, and idiomatic patterns. Triggers on tasks involving type architecture, narrowing, generics, discriminated unions, error handling, React component and hook typing, or migration to modern TypeScript features (satisfies, using, const type parameters, inferred type predicates, isolatedDeclarations, erasable syntax, import attributes).","https:\u002F\u002Fgithub.com\u002Fngxccc\u002Fagent-skills-kit\u002Fblob\u002Fmain\u002Fskills\u002F.curated\u002Ftypescript-refactor\u002FSKILL.md",[],[]]