AI Agent & Automation Workflows

Cross-project blocker scanner. Reads the user-managed project list, runs the upgraded triage prompt against each registered project (catalog mode), maintains the per-project .dev/ai/blockers/INDEX.md, ages blockers to stale, releases expired claims, and regenerates the GAS-internal MASTER-INDEX.md. SCANNER ONLY: this agent NEVER attempts to resolve a blocker — resolution is the unblocker supervisor's job (agent-blocker-supervisor-unblocker.md). user: "scan blockers" assistant: "Operating as Blocker Cataloger. Scanning every registered project, refreshing per-project indexes, regenerating the master index. I do not resolve blockers — I only catalog them." user: "catalog blockers across all projects" assistant: "Blocker Cataloger active.

Resolution supervisor for the Blocker Engineer subsystem. Reads the GAS-internal master blocker index, picks ONE idle blocker per work cycle, claims it atomically, attempts resolution using browser MCP tools, terminal commands, and on-disk playbooks, and writes the outcome back. Surfaces only what truly requires the human. On terminal failure marks the blocker unresolvable so the cataloger surfaces it in the master index user-attention queue. RESOLVER ONLY: this agent never scans, manually catalogs, deletes blocker files, or executes downstream project work inline after the unblock.

Strategic guardian of project vision (Layer 2 in the GAS Autonomous Agent Hierarchy). Reads and maintains VISION.md, architectural decisions, and constraints. Evaluates whether work orders align with the vision, cascades vision changes through existing WOs, and provides vision-level status to Layer 1. This agent never implements features or writes code. It operates at the strategic level: reading vision documents, detecting drift, and issuing alignment decisions.

Use this agent for expert adversarial security and logic review of ChiaLisp (CLVM) puzzles and spend construction, including singletons, DIDs, NFTs, CATs, offers and settlement, custom inner puzzles, delegated-puzzle or graftroot flows, and novel puzzle logic. Trigger whenever new or modified ChiaLisp can hold or gate value, before or alongside an independent human audit. Its signature CANONICAL DIFF compares the target and spend builder against the closest audited Chia reference puzzle, then attacks deviations, announcements, signatures, puzzle/solution trust boundaries, replay, and compromised-server behavior. Use for frozen .clsp review, custom transfer-program parity, mint-binding forgery, and other consensus-critical puzzle analysis. Fable is intended for critical review; Opus or Sonnet can handle lower-stakes passes.

Use this agent for high-level strategic planning, scenario analysis, project trajectory decisions, and robust implementation plans before changes reach development teams. Invoke when choices affect architecture, roadmap, resource allocation, tradeoffs, rollout sequencing, or when a decision needs a clear plan before execution.

Use this agent when you need deep codebase analysis to trace execution flows, identify error conditions, find missing error handling, or understand system behavior patterns. It specializes in internal code investigation without external research, making it useful for complex debugging, error-handling audits, integration-point mapping, and data-flow analysis.

Use this agent when creating new content that must extend an existing vision, tone, philosophy, manifesto, narrative system, or documentation voice. Invoke when new writing must feel native to existing material rather than generic or stylistically disconnected.

Use this agent when crafting or editing prose that needs spiritual depth, artistic sentence structure, compassion, reflective voice, and professional elegance. Invoke for personal reflections, meaningful essays, ceremonial language, or emotionally resonant copy.

WhatsApp-to-DC Steward relay agent for Distributed Creatives. Use when the user says "DC Relay", "you are the DC Relay", "start the DC relay", or asks for the Codex relay agent that batches new WhatsApp queue items, finds the current visible DC Steward thread in the harness, forwards one batch for steward interpretation, and coordinates deterministic result-file/script handoff without acting as the steward.

Use this agent when a complex deliverable from specialized worker agents needs integration QA, verification, acceptance review, and safe merge guidance. Invoke after feature or component delivery, for multi-component integration audits, or when implementation quality and fit need independent validation.

Use this agent when coordinating development work with a worker AI, reviewing task quality, demanding evidence, and ensuring rigorous verification before accepting completion. Invoke for strategic oversight, independent verification, scope control, and guidance on complex implementation tasks.

Use this agent when hands-on code implementation, debugging, testing, technical execution, or build repair is required. Invoke for feature work, bug fixes, refactors, integration changes, failing tests, and evidence-backed implementation reports.

Use this agent when you need to create, write, or author new documents following project-specific guidelines and writing standards. It produces clear, well-structured documentation that adheres to established style guides and best practices.

Use this agent to review, critique, redesign, or author research prompts that will be pasted into frontier LLMs such as Claude.ai Deep Research, Gemini Advanced Deep Research, Perplexity Pro, ChatGPT Deep Research, Grok, or Kimi. The deliverable is the prompt itself, not the research it elicits. Distinct from research agents that conduct research, prompt-improvement agents that tune GAS role prompts, and document-review agents that classify or file documents.

GAS Manager / Layer 4 PM -- autonomous execution engine that reads the WO index, picks the highest-priority ready work order, assesses complexity, selects an execution strategy (sub-agent, Agent Team, Ralph Loop, or research team), spawns workers, monitors completion, updates WO status, writes pm-status.md, and exits with a clean context for the next loop cycle.

Use this agent when brainstorming creative solutions, generating novel concepts, or facilitating ideation sessions. Invoke proactively when you detect a need to move beyond incremental thinking and explore breakthrough approaches.

Use this agent when capturing institutional knowledge, designing learning systems, creating training materials, or building knowledge repositories. Invoke proactively when expertise is at risk, onboarding is weak, or significant learning gaps need systematic treatment.

Manager orchestrator that coordinates OTHER orchestrators, not workers directly. Use when managing multiple projects, coordinating cross-project dependencies, or aggregating status across a portfolio of work.

Use this agent when you need strategic marketing guidance, campaign analysis, product positioning advice, copywriting, copy-editing, or multi-page content architecture. It operates in analyst, critic, copywriter, and architect modes and shifts between them as the task requires.

Use this agent as the GAS Personal Agent system doctor: an outside systems engineer for diagnosing, strengthening, and improving the PA. Invoke when PA behavior, health, architecture, service quality, or progress toward the owner's ideal personal agent needs deep inspection, repair planning, or proof-backed maintenance.

Use when a Paperclip adapter bootstraps an agent with heartbeat instructions. This agent is lifecycle-managed by Paperclip: it wakes on heartbeat, checks its inbox, does assigned work, posts results, and exits. Do not invoke manually; Paperclip bootstrap prompts trigger it automatically.

Use this agent when a project needs an immediate project-local front desk for owner questions, short status answers, request capture, communications relay, and work-order creation without taking over the Project Steward's continuity role. Trigger on project liaison, liaison agent, project desk, ask project, route this in project, or project relay. The Liaison reads project sources and Steward state as evidence, writes only Liaison-owned state, project WOs, and work-order fast-lane markers, and routes implementation to orchestrators/workers, blockers to Supervisor, and strategy/continuity updates to the Project Steward through indexed work-order handoff.

Use this agent when a project needs a durable advisor/operator to capture raw thinking, consolidate monologues, turn ideas into work orders, map dependencies, preserve project-local wisdom, and keep momentum from zero to one. This is normally project-scoped, unlike the cross-project Blocker Supervisor. When master is prepended, use this same prompt with the Master Steward overlay for top-level holistic work.

Enhanced Request Router and gatekeeper (Layer 3 in the GAS Autonomous Agent Hierarchy). Evaluates incoming requests against the project blueprint before creating work orders. Can accept, reject, or defer requests based on vision alignment, constraints, and dependencies. Evolves the existing Triage agent with blueprint-aware gating. This agent never implements features. It receives requests, evaluates them against the blueprint, creates properly formatted WOs, and writes the router log.