geas-grimoire

Execute the technical plan for a feature using bounded, task-by-task execution. Acts as the "Goldfish", relying on strict constraints from the "Elephant" (spec.md).

Define requirements, specifications, architecture, and step-by-step task plans for a new feature or modification. Orchestrates the creation of the Elephant (spec.md) and Goldfish Checklist (plan.md).

Establish a project governance document (constitution.md) with immutable architectural articles that all Bolts inherit. Run once per project before bolt-roadmap.

Investigate and document technology choices before architecture is locked. Invoked from bolt-intent Phase 2 when new technology decisions are required. Produces research.md. Optional — skip with documented reason when Bolt extends an established pattern.

Structured post-Bolt retrospective that captures process improvements, proposes constitution amendments, and updates bolt-* skills. Targets the development process — not the project. Distinct from bolt-synthesis.

Create a comprehensive strategic roadmap and diagram for the project by decomposing high-level goals into atomic, sequential/parallel bolts.

The Warden. Run the two-phase Pre-Flight Check (Deterministic Verify + Semantic Audit) against the completed implementation to ensure no architectural drift occurred.

Manage the project's knowledge base. Run post-audit to clean up transient Goldfish checklists (plan.md), merge feature updates back into master design docs, and promote architectural decisions to global project knowledge.

Programmatically extract metadata from YML-based data contracts (dbt schema.yml, metadata.yml, *_contract.yml, *_schema.yml, etc.) to infer column names, data types, semantic descriptions, nullability, tags, and test definitions — without reading entire files into context. Use this skill whenever you encounter a task involving data contracts, dbt models, table validation, schema documentation, or query generation from structured YML sources. Trigger autonomously when you discover relevant YML files during work, even if the user has not explicitly asked for contract extraction. Token-efficiency is a core goal: extract only what is needed for the task at hand.

Build, critique, or optimize prompts for Claude — whether for Claude.ai chat, API system prompts, or Claude Code CLAUDE.md / slash command / agent definitions. Use this skill whenever the user wants to write a new prompt from scratch, improve an existing one, reverse-engineer a reference output into prompt rules, or diagnose why a prompt isn't working. Trigger on: "write me a prompt", "improve my prompt", "help me prompt Claude", "my Claude prompt isn't working", "create a system prompt", "write a CLAUDE.md", "optimize this instruction", "turn this into a prompt", or any request to craft instructions for Claude in any context. Always use this skill before writing any substantial prompt — even if the user hasn't explicitly asked for "a skill".

Systematically wrap up a session by summarizing progress, scouting for reusable skills, and cleaning up context.

Create professional Word documents (.docx) from company templates. Use when: (1) Creating reports, proposals, or contracts, (2) Generating documents with charts and data analysis, (3) Drafting content that needs company styling. Agent writes narrative content based on user-provided data/context.

Build and manage a persistent knowledge base (second brain) using Markdown files. Use when you need to store, retrieve, or link complex concepts, project context, or 'remember' information across sessions. Triggers on: (1) 'Remember this', (2) 'Build a knowledge base', (3) 'Document this concept', (4) Summarizing findings into long-term memory.

Expert AI Assistant for translating dense academic algorithms, pseudocode, and mathematical formulas into robust, numerically stable, and hardware-agnostic production code pipelines.

Interface with the Google Jules CLI to delegate asynchronous coding tasks. Use this tool to execute a specific plan, start a background session, or retrieve results from a remote Jules agent.

Build and manage a persistent knowledge base (second brain) using Markdown files. Use when you need to store, retrieve, or link complex concepts, project context, or 'remember' information across sessions. Triggers on: (1) 'Remember this', (2) 'Build a knowledge base', (3) 'Document this concept', (4) Summarizing findings into long-term memory.

Analyzes code for performance, maintainability, quality, and architecture

Interface with the Opencode CLI to execute AI-driven coding tasks, generate code snippets, and manage agent sessions non-interactively. Use this when the user asks to "ask Opencode" or "use Opencode" to perform a task.

Detects, analyzes, and suggests changes to the Opencode CLI configuration file (config.json). Use this when the user complains about Opencode performance, asks to change the default model, or wants to modify CLI behavior.

Manage and synchronize agentic skills from public GitHub repositories using the `skm` CLI utility. Use this skill when you need to: (1) Find and fetch new skills from external repositories, (2) Keep existing skills updated, (3) List available skills in the marketplace, or (4) Bootstrap a new environment with a specific skill set.

Analyzes the recent conversation history or a completed task to detect reusable patterns. Generates a new SKILL.md file to automate that workflow in the future. Use this when the user asks to "create a skill from this," "save this workflow," or "suggest a new skill."

Enforces Test-Driven Development methodology

Guide for creating effective, structured, and token-efficient technical documentation. Use when taking on documentation tasks, defining system architecture, writing technical specs, or whenever the user asks for "structured documentation".