standardbeagle

Record + replay an interactive guided demo overlay on the live app — self-advancing walkthrough, labelled gesture affordances. 瀏覽器互動導覽。 Use when: demo a feature, create walkthrough, run walkthrough tool, guided product tour, replay a flow, show what was built, onboarding tour, step-by-step demo overlay.

Adversarial cooperation loop for code quality verification with plan adjustment at each phase. 對抗協作循環：實施者與驗證者互相制衡，逐階調整計劃，確保代碼品質。 Use when: execute dart task, run quality pipeline, adversarial review, implement with verification

Search + synthesize prior agent session history. Claude Code primary (~/.claude/projects/); Codex/Cursor secondary. Discover -> branch/keyword filter -> extract skeleton/errors -> session-historian subagent. Use when: what was tried, prior attempts, past decisions, recent work, why approach abandoned, cross-session context, even without saying 'sessions'.

Canonical verdict-only return schema for adversarial reviewer subagents. Reviewers emit compact ≤30-line verdict block; main loop parses verdict + blockers only. Use when: authoring or updating reviewer agent, building verdict consumer, debugging reviewer report parsing.

Diagnose + configure compound-engineering env — CLI deps, plugin version, repo-local config. Guided install for missing tools, gitignore + config bootstrap. Use when: troubleshoot missing tools, verify CE setup, onboard repo, refresh config template, /ce-setup re-check. Skip: unrelated to compound-engineering tooling.

Resolve review feedback by evaluating validity and fixing issues in parallel. Pre-commit-first. Use when: applying review feedback from local reviewer agent output, resolving pre-commit review findings, working through code-review agent suggestions, addressing reviewer-flagged issues before commit. Also: PR comments via gh as secondary path. 解決審查回饋：本地審查代理輸出優先，PR次之。

Multi-source research recipe via caveman-mcp — gather URLs/PDFs/PRs in parallel, compress each, synthesize with provenance. Pairs with knowledge-hygiene for conflict detection. 多源研究：並行採集、壓縮、含出處合成。 Use when: research topic across sources, prep literature review, ingest reading list, build evidence pack, multi-source synthesis with citations

"Start Ralph Wiggum adversarial cooperation loop on dartboard with plan adjustment. 在看板上啟動Ralph Wiggum對抗合作循環，含計劃調整。 Use when: start execution loop, run dartboard tasks, begin adversarial loop, automate task execution, process task queue"

You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.

Use when an HTML UI beats terminal text — collecting structured answers (decision trees, multi-select, code/SQL fields), showing visual comparisons or mockups, running interactive demos, or presenting decisions for approval. A local browser companion renders markdown+YAML screens and streams the user's answers back. Use when: many questions at once, layout/visual choices, mockup or diagram feedback, config wizard, demo review, approve/revise decisions. Skip: a single quick question (use AskUserQuestion or plain text).

Use when you have a spec or requirements for a multi-step task, before touching code

Use when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions

Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies

Use when you have a written implementation plan to execute in a separate session with review checkpoints

Use when implementation is complete, all tests pass, and you need to decide how to integrate the work - guides completion of development work by presenting structured options for merge, PR, or cleanup

Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation

Run the release-blocking Pester gate (tests/PsBash.Tests.ps1) the reliable way — build cmdlets, refresh beside-module DLLs, isolate PSModulePath from a stale installed PsBash, run Invoke-Pester.

Use (or extend) the shared PsBuild PowerShell-text builder and the OS-interface FileSystemHelpers/RunChildProcess helpers instead of hand-rolling PS strings or platform branches

Add a new bash command to the ps-bash transpiler and runtime

Add new grammar production to the bash parser (new token, AST node, or syntax)

Debug a bash command that produces wrong output or crashes in ps-bash

Rebuild and publish the local ps-bash exe for testing in opencode

Audit one bash feature against the QA rubric and write its section into docs/testing/interactive-parity-audit.md

Validate specs against source code (detect drift)

Coordinate alert/event delivery to AI agents — single-queue rule, layered gates (dedup, batch, activity-defer, outage-hold), and the forbidden parallel-path pattern that causes spam

Diagnose autostart, proxy, and process issues — why things aren't starting, connecting, or recovering

Design and implement health check, doctor command, and reconciliation logic for the daemon

Navigate the agnt codebase — find where features live, how subsystems connect, and which files own what

Cut a tman release — pick the version from what actually changed, update CHANGELOG and every doc surface that drifted, review as an outside consumer would, verify the real binary, then tag and push. Use when asked to release, cut a version, publish, or bump tman.