agentboard

Quick progress snapshot of a workspace board — card counts, blocked items, recent activity

Interactive spec-building session — brainstorm work, produce a spec document, and create workspace board cards

Onboard an agent to AgentBoard — verify connectivity and create or select a project

Run the workspace orchestration pipeline — planning, review, implementation, and audit waves with parallel subagents

Resume AgentBoard work — read the skill, check server health, find active project, and claim next task

> **This file is no longer the active prompt for Wave 4.**

Phase B of audit pipeline. Reads the pre-gathered audit facts bundle from audit-research-agent and writes a rigorous audit report with PASS/PASS WITH NOTES/FAIL verdict. Full Expert Standard process. Does not modify source files. Invoke from the workspace-orchestration skill — the orchestrator passes card_id, board_id, agent_id, card_title, and audit_facts_bundle in the prompt.

Phase A of audit pipeline — mechanical fact-gathering from git diff, codegraph, and existing artifacts. Produces a structured AUDIT_FACTS_BUNDLE_V1 artifact for audit-compose-agent. Does not evaluate quality or write the audit report. Invoke from the workspace-orchestration skill — the orchestrator passes card_id, board_id, agent_id, and repo_root in the prompt.

Wave 3 of AgentBoard workspace orchestration. Executes the most recent `plan` artifact on a workspace card — writes code, modifies files, runs build/lint — and submits an `implementation_note` artifact. Invoke from the workspace-orchestration skill — the orchestrator passes card_id, board_id, agent_id, and card_title in the prompt.

Phase B of planning pipeline. Reads the pre-gathered facts bundle from planning-research-agent and writes a rigorous, audit-grade implementation plan. Full Expert Standard process, Clear Thought reasoning, Context7 verification, and Gate A/B/C compliance — without the codebase discovery phase (handled by planning-research-agent). Invoke from the workspace-orchestration skill — the orchestrator passes card_id, board_id, agent_id, card_title, and facts_bundle in the prompt.

Skill for using the AgentBoard AI project management app and its cloud-hosted MCP server. Use when the user wants to manage phase-based projects (phases, milestones, documents, tasks), work with workspace boards (apps, boards, cards, artifacts) for ad-hoc orchestration, authenticate the AgentBoard MCP, interact with any AgentBoard MCP tool, or needs guidance on either workflow. Triggers on "manage project", "use agentboard", "agentboard workflow", "create a project in agentboard", "set up agentboard", "authenticate agentboard", "create a workspace card", "create a board", "orchestrate cards", "submit artifact", "workspace board", or "fetch a card".

Always-on semantic search over the current project. Use it before editing unfamiliar code, when looking for callers/callees, when checking how a similar feature is already implemented, or when you need to know what depends on a file before changing it.

Systematically discover quality issues in an existing codebase through a single broad sweep, then organize findings into an actionable document. Use when the user wants to clean up a vibe-coded app, audit codebase quality, find technical debt, survey an app for problems, or prepare an existing codebase for improvement. Triggers on "sweep my codebase", "clean up this app", "audit code quality", "find issues in this project", "this app is a mess", "technical debt", "vibe coded", "what's wrong with this code", "review my project", or any request to assess or inventory problems in an existing codebase before fixing things. Use this skill even if the user doesn't explicitly say "sweep" — any request to find or catalog problems in a codebase qualifies.

The foundational evaluation frame for all engineering work. Activates whenever Claude is making an engineering judgment of any kind — writing code, reviewing code, debugging, making architecture decisions, assessing quality, evaluating completeness, refactoring, choosing between approaches. Especially activates when Claude is about to approve or praise something: 'looks good', 'well-structured', 'correctly implemented', 'this is fine', or any positive quality judgment. If Claude is producing or evaluating engineering work, this skill applies — even for small tasks like writing a helper function or fixing a bug. This skill changes how Claude thinks, not what it delivers. For a structured deep review with findings and classifications, use the /expert-review command instead.

Use when orchestrating parallel subagents across workspace board cards for planning, review, implementation, and audit waves

4 hooks across 3 events