AI Agent Workflow Collection

Create or revise high-level flow diagrams for client or non-technical audiences. Use when the user needs a workflow, process, integration, or system-interaction diagram that communicates business-level behavior without exposing implementation details.

Hand a drafted artifact to the user through the clipboard, with post-copy verification and on-demand re-copy, so the user can review it and paste it into a destination the agent must not or cannot write to directly (a review-gated personal vault such as Obsidian, a web form, another application). Trigger phrases include "copy for me", "幫我複製", "pbcopy", "過目再貼". Do NOT use when the agent can write the content into the target repo directly, or for multi-file transfers.

Scan the current session for candidate CLAUDE.md/AGENTS.md baselines or candidate Claude Code skills, and write them as structured notes to a configurable output folder for later manual review. Use when the user asks to extract improvement ideas, capture process improvements, or save candidate baselines/skills surfaced during this session — phrases like "extract improvements", "capture what we'd add to CLAUDE.md", or "save this as a candidate skill idea".

Evaluate an external article, repository, tool list, paper, framework, or workflow before promoting it into repo artifacts. Use when deciding whether outside methodology material should become a Rule, Skill, Context term, ADR, Spec, Issue, or No-op.

Query work items and pull requests from an Azure DevOps organization. Use when asked to pull tickets, stories, support requests, PRs, or activity summaries from Azure DevOps. Encodes the correct tech-path discovery sequence so you do not waste turns on dead ends.

Chain the personal end-of-session skills — a session-log capture skill, lesson-extraction, and improvement-extraction — into a single wrap-up pass instead of invoking them one at a time. Use when the user wants to close out, wrap up, or end a work/coding session and capture everything worth keeping in one go. Triggers on "wrap up this session", "close out", "let's wrap", "session closeout", "收工", "收官三連".

Create a bounded manual execution packet for large AI-agent team workflows, or refuse/downgrade when the task should stay single-agent. Use when work is multi-domain, parallelizable, and context-heavy enough to need role selection, context packets, ownership boundaries, handoffs, budgets, verification, package-execution policy, and external-system credential policy before launching workers.

Turn user-provided short videos, video collections, photo collections, or mixed media into a social-platform Live Photo card when one small motion point explains more than a static image, with collection screening, first-frame, duration, crop, and publish-path checks.

Read-only engineering review with a fixed findings/verdict output contract. Use when the review needs that contract, a review context bundle, or a non-GitHub diff source (e.g. Azure DevOps, a patch file, a bounded set of repos and docs). For routine review of a local diff or GitHub PR, prefer the built-in /code-review or /review instead.