Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/davidbuenov/dbv-specs-ops/llms.txt

Use this file to discover all available pages before exploring further.

The entire dbv-specs-ops framework is a collection of plain Markdown template files — there is no binary to install, no package manager to run, and no build step required. You get the framework by copying a folder. Each file serves a specific role in the Spec-Driven Development lifecycle: some establish the rules the AI must follow, some capture decisions and state across sessions, some are produced as phase outputs, and some activate the framework on a specific AI platform. Understanding this file map is the fastest way to gain intuition for how the system works as a whole.
All paths below are relative to the root of your project (where you placed the framework files). The docs/ folder holds immutable reference and specification documents; the root holds living operational files that the AI reads and writes throughout development.

Comprehensive file reference

FileLocationRead byCreated whenPurpose
MASTER_PROMPT.mddocs/AISession start (manual platforms)Core rules and constraints. Defines the 6-phase SDD workflow, context management strategy, security boundaries, and development standards the AI must follow. On auto-loading platforms, equivalent rules are delivered via CLAUDE.md, GEMINI.md, or .windsurfrules.
project.config.mdrootAIProject bootstrap (template provided)Project identity: name, author, license, language list, Agent Readiness flag, and file-header template. Also contains the Model Routing Guidelines table. The AI fills in placeholders during the bootstrap interview.
SPECIFICATIONS.mddocs/Both/spec phaseThe “What” and “Why”. Problem statement, user profiles, key scenarios, functional requirements, out-of-scope items, risks, open questions, and Evals criteria. The AI will not proceed to /plan until open questions are resolved.
ARCHITECTURE.mddocs/Both/plan phaseThe “How”. Tech-stack table, directory structure, key technical decisions (auth, state, style), external integrations, and the Agent Harness configuration (MCP servers, sandboxing, guardrails, Agent Readiness architecture).
DESIGN.mddocs/Both/spec phase, if UI existsThe “Look”. Visual design system: color tokens, typography scale, spacing, and UI component definitions. Created only for projects with a user interface. Optional for pure API or CLI projects.
memory.mdrootAIFirst /plan or /build sessionQualitative persistent context. Stores active context, Architecture Decision Records (ADRs), lessons learned, and a relations map. The AI must consult this at the start of every session to recover technical continuity across token-limit boundaries.
task.mdrootAIFirst /plan sessionOperational state. Quantitative progress checklist, backlog items, technical-debt log, and Context Snapshots that let the AI resume work after a session ends.
CLAUDE.mdrootAI (Claude only)Template providedPlatform activation file for Claude Code (CLI, VS Code, Desktop) and Cursor. Automatically loaded at session start, delivering the same rules as MASTER_PROMPT.md without manual pasting.
GEMINI.mdrootAI (Gemini / Antigravity)Template providedPlatform activation file for Gemini CLI and the primary auto-load trigger for Antigravity (Google DeepMind VS Code extension). Automatically loaded at session start.
ANTIGRAVITY.mdrootAI (Antigravity)Template providedExtended behavior configuration specific to the Antigravity platform, including optional Knowledge Index (KI) setup and Antigravity-specific planning artifact conventions.
.windsurfrulesrootAI (Windsurf)Template providedPlatform activation file for the Windsurf IDE. Automatically loaded at session start; contains the same enforcement rules as MASTER_PROMPT.md formatted for Windsurf’s rule loader.
ADOPTION_PROMPT.mddocs/AIOn demand (existing projects)Onboarding guide for adopting SDD in a project that already has code but no specifications. The AI reads this when instructed to analyze an existing codebase and generate SPECIFICATIONS.md, ARCHITECTURE.md, and task.md from scratch.
UPGRADE_PROMPT.mddocs/AIOn demand (version upgrades)Zero-collision upgrade agent. Instructs the AI to detect the current framework version from project.config.md, download only the changed framework files, and apply them without touching source code, specs, or changelog.
AGENTIC_ENGINEERING.mddocs/HumanTemplate providedHuman-facing reference document. Explains the theoretical foundations of Agentic Engineering: Conductor vs. Orchestrator modes, Evals, the Agent Harness, and the transition from interactive prompting to autonomous codebase-factory operation.
CHANGELOG.mdrootBoth/build phase (first entry)Version history in Keep-a-Changelog format. The AI writes to the [Unreleased] section during /build and /test, and promotes entries to a versioned section with a date during /ship.
README.mdrootBoth/ship phaseHuman-facing project overview. Generated from README.template.md during bootstrap (template is then deleted). Updated by the AI on every /ship to reflect the current stable version and feature set.
walkthrough.mdrootBoth/ship phase (AI writes)Feature walkthrough. A plain-language summary of what was built, tested, and delivered in the current release cycle. Written by the AI at the end of each /ship phase.
implementation_plan.mdrootAI writes, Human approves/plan phase (complex tasks only)Detailed execution plan for complex tasks (those touching more than 3 files, involving auth/payments/sensitive data, or estimated at more than 150 new lines). Includes a YAML frontmatter block with dependencies, risks, and rollback_strategy. The AI waits for explicit human approval before proceeding to /build.

Minimal required vs. generated on demand

Not all files need to exist before you start. The framework distinguishes two tiers: Minimal required set — these files must be present at session start for the AI to load context and enforce the SDD workflow correctly:
  • project.config.md (project identity and routing rules)
  • CLAUDE.md / GEMINI.md / .windsurfrules (one platform activation file matching your AI tool)
  • docs/MASTER_PROMPT.md (required for manual-load platforms like ChatGPT or Gemini Web)
If the platform activation file is missing or empty, the AI will behave as a generic assistant and will not enforce the SDD lifecycle. Always verify the correct activation file is present for your platform.
Generated on demand — these files are created by the AI during the appropriate phase and should not be pre-populated manually:
  • docs/SPECIFICATIONS.md → created during /spec
  • docs/ARCHITECTURE.md → created during /plan
  • docs/DESIGN.md → created during /spec, only if the project has a UI
  • task.md → created during the first /plan session
  • memory.md → first written during /plan or /build when a decision is logged
  • implementation_plan.md → created during /plan for complex tasks only
  • walkthrough.md → created during /ship
  • CHANGELOG.md → first entry added during /build
  • README.md → generated from README.template.md during bootstrap

Files grouped by category

Core Rules

Files that define how the AI must behave. These are read at session start and establish the enforcement layer for the entire SDD workflow.
  • docs/MASTER_PROMPT.md — full rule set (manual platforms)
  • project.config.md — project identity + model routing
  • docs/AGENTIC_ENGINEERING.md — human reference for agentic principles
  • docs/ADOPTION_PROMPT.md — on-demand SDD adoption for existing projects
  • docs/UPGRADE_PROMPT.md — on-demand zero-collision framework upgrade

Context Files

Files that persist knowledge across sessions, eliminating AI amnesia and enabling long-running projects without context loss.
  • memory.md — qualitative decisions and ADRs
  • task.md — quantitative progress and context snapshots
  • docs/SPECIFICATIONS.md — requirements and acceptance criteria
  • docs/ARCHITECTURE.md — technical decisions and harness config
  • docs/DESIGN.md — visual design system (UI projects only)

Phase Outputs

Files that are written or updated by the AI as it moves through the six development phases.
  • implementation_plan.md/plan output (complex tasks)
  • CHANGELOG.md — updated during /build, /test, and /ship
  • walkthrough.md/ship output
  • README.md — updated on every /ship

Platform Activations

Files that trigger automatic context loading for a specific AI platform. Use only the file that matches your tool — the others are optional.
  • CLAUDE.md — Claude Code, Cursor
  • GEMINI.md — Gemini CLI, Antigravity
  • ANTIGRAVITY.md — Antigravity extended config
  • .windsurfrules — Windsurf IDE
When adopting dbv-specs-ops into an existing project, use Option B (subfolder isolation): copy the entire dbv-specs-ops folder into your repo root and place a small activation file at the root that points the AI to dbv-specs-ops/docs/MASTER_PROMPT.md. This prevents the framework files from overwriting your existing README.md, CHANGELOG.md, or source code.

Build docs developers (and LLMs) love