dbv-specs-ops Version History: v1.0.0 through v2.1.0

​

Added

• Agent Readiness integration: Native support in the framework for web projects and public APIs. The development cycle now guides the creation of:
  • robots.txt with exclusion directives and Content-Signals (ai-train=no, search=yes, ai-input=yes).
  • Semantic navigation map (llms.txt and auth.md for anonymous/OAuth registration flows).
  • OIDC/OAuth discovery metadata and signatures in .well-known/ (api-catalog, oauth-protected-resource, oauth-authorization-server, and http-message-signatures-directory).
  • Bot and MCP server identity cards (agent.json and mcp.json).
  • Skills structure and repository (agent-skills/ with an index.json index and individual SKILL.md guides).
  • Markdown content negotiation mechanisms (Accept: text/markdown) and HTTP Link headers.
• Template updates:
  • project.config.md: Agent Readiness (Web) configuration field added to project identity.
  • docs/SPECIFICATIONS.md: Integrated checklist in the technical proposal section and a new AI context-consumption risk entry in section 6.
  • docs/ARCHITECTURE.md: External Agent Interface section added to the Agent Harness template.
• SDD lifecycle updates (MASTER_PROMPT.md):
  • /spec phase evaluates Agent Readiness viability.
  • /build phase guides structured file creation.
  • /ship phase verifies injected Link headers.

​

Changed

• README.md: Added detailed description of the framework’s main features in both English and Spanish.

​

Fixed

• README.md: Corrected a contradiction in the Quickstart where /plan was suggested as the first command instead of /spec.

​

Added

• Agentic Engineering: Full integration of concepts from the Google white paper “The New SDLC With Vibe Coding”.
• Implicit Work Modes: Automatic and invisible classification between Conductor mode (interactive IDE short loops) and Orchestrator mode (asynchronous background execution) during the /plan phase.
• Unified Evals in /test: The /test phase now covers both classic deterministic tests and probabilistic Evals (output rubrics, trajectory evaluation, and hallucination verification).
• Security Review in /code-simplify: Mandatory audit of AI-generated code to prevent false dependencies (slopsquatting), secret injection, and vulnerable inputs.
• Agent Harness: Transition from the old MCP section to an Agent Harness specification in docs/ARCHITECTURE.md (defining static vs. dynamic context, sandboxing, and security hooks).
• Foundation guide: New docs/AGENTIC_ENGINEERING.md document detailing the theoretical reasoning behind v2.0.0.
• Model Routing Guidelines: Optimal AI model assignment directives in project.config.md to reduce operational cost (OpEx/Token Burn).
• Active MCP and Skills suggestions: Mandate in /spec and /plan to propose the creation of local MCP servers and dynamic skills in projects.
• Upgrade Prompt updated: docs/UPGRADE_PROMPT.md updated to include v2.0.0 so existing projects can upgrade automatically.

​

Added

• Trust Boundary (<trust_boundary>): New section in MASTER_PROMPT.md that explicitly declares the separation between valid directives (XML tags in the prompt) and project data (SPECIFICATIONS.md, task.md, memory.md). Prevents silent execution of imperative text injected into data files.
• Granular Memory Triggers: In /plan, if the Adversarial Review consciously accepts a risk, it must be immediately recorded in memory.md. In /build, if ARCHITECTURE.md is contradicted. In /test, if a test invalidates an assumption in SPECIFICATIONS.md. No need to wait until /ship.
• Maintenance policy in memory.md: New 🧹 Maintenance Policy section with the goal of keeping the file below 200 active lines, consolidation instructions, and use of memory.archive.md.

​

Fixed

• Anti-boilerplate Adversarial Review: The <adversary> block must now cite at least one concrete noun present in docs/SPECIFICATIONS.md. Generic responses such as “what if the network fails?” are eliminated.
• Bootstrap without auto-git-execution: The bootstrap flow no longer executes git init automatically. It displays the command and requests explicit user confirmation.
• memory/task boundary clarified: memory.md now explicitly distinguishes that task.md is operational (exact next step) and memory.md is thematic (qualitative context, product area in focus).

​

Fixed

• English documentation inconsistency in README.md: The “Adopting an Existing Project” section described the old interview flow (“6 questions, one at a time”). Updated to reflect the new draft-with-assumptions flow.
• Inconsistency in project.config.md: The file description callout said “the AI will ask 3 quick questions”. Updated to reflect the real flow (full draft with [ASSUMPTION: ...]).
• Internal evaluation files in .gitignore: Added docs/EvaluacionChatGPT.md, docs/EvaluacionCopilot.md, and docs/EvaluaciondeClaude.md to .gitignore so they are not versioned.

​

Added

• Enforcement Layer (Hard-Law): Transition from suggested methodologies to strict contracts.
• Phase Gates in /plan: implementation_plan.md now mandatorily requires a YAML Frontmatter defining dependencies, risks, and rollback strategy.
• Memory Gate in /ship: Mandatory step added to print <memory_update_proposal> XML before closing tasks, forcing persistence of qualitative context in memory.md.
• Adversarial Review (/plan): Implemented a forced internal debate (<builder> vs <adversary>) in XML format to evaluate edge cases before the AI breaks down tasks.

​

Fixed

• Cognitive redundancy: Extreme DRY refactoring of adapters (CLAUDE.md, GEMINI.md, .windsurfrules, .github/copilot-instructions.md, ANTIGRAVITY.md). Duplicated rules about project state, lifecycle, and bootstrap have been removed. They now act as lightweight pointers to docs/MASTER_PROMPT.md to reduce cognitive load on the LLM.
• Initial interview friction: The bootstrap and adoption process no longer asks questions one by one. The AI now proposes a complete draft with marked assumptions ([ASSUMPTION: ...]) that the user validates or corrects in a single step.
• Template cleanup: task.md restored to a clean template state (previously carried over meta-history from the framework).
• Consistency: Added memory.md to the initial reading tables of all adapters.
• Noise prevention: Added explicit instructions in memory.md for the AI to delete examples when creating real content.

​

Added

• memory.md: New standard file to separate qualitative context (technical decisions, lessons learned, relationship map) from quantitative progress (task.md), preventing architectural drift and AI context loss.
• Architect Review in /plan: When invoking /plan, the AI now performs a prior validation assuming the role of Software Architect to detect edge cases and logical vulnerabilities before breaking down tasks.

​

Changed

• XML Prompts: docs/MASTER_PROMPT.md, docs/ADOPTION_PROMPT.md, and docs/UPGRADE_PROMPT.md have been completely rewritten using semantic XML tags (<workflow>, <boundaries>, etc.) to improve compliance in models such as Claude 3.5, Gemini 1.5, and GPT-4o.
• README.md: Updated to include memory.md in the file structure and the Architect Review in the /plan phase definition.
• GEMINI.md: Updated to include reading of memory.md at session start.
• .gitignore: Updated to ignore implementation_plan.md and walkthrough.md.

​

Added

• docs/DESIGN.md: Visual design system template (YAML design tokens + prose) following the design.md standard by Google Labs. Includes native Dark Mode support as an original contribution to the standard.
• docs/UPGRADE_PROMPT.md: Framework upgrade agent: detects the project’s current version, calculates the change delta, downloads framework files from GitHub (with manual fallback if there is no network), and never touches project files.
• framework_version in project.config.md: New field to track which framework version was last used to initialize or update the project.

​

Changed

• GEMINI.md, CLAUDE.md, ANTIGRAVITY.md, .windsurfrules, .github/copilot-instructions.md: All updated to read docs/DESIGN.md at session start (if it exists) and mention the design system in the adaptation section.
• docs/MASTER_PROMPT.md: /spec phase updated — if the project has UI and docs/DESIGN.md does not exist, create it in this phase.
• docs/SPECIFICATIONS.md: Section 4 links to docs/DESIGN.md when the project has a user interface.
• README.md: New “Upgrading an Existing Project” (EN) and “Actualizar el Framework” (ES) sections with single-download flow for UPGRADE_PROMPT.md.

​

Added

• README.template.md: Base skeleton for user projects.
• Bootstrap Language question: A 5th question added to the initial interview to choose the project’s documentation language (ES, EN, or Bilingual).
• Project README auto-generation: The AI now generates a customized README.md for the user project using the template and then deletes the .template file.
• Reference section for phases: Detailed table in the main README explaining the 6 development phases (/spec, /plan, /build, /test, /code-simplify, /ship) and their trigger commands.

​

Added

• project.config.md: New root file for project identity: name, author/company, license and file header templates for all languages (JS, Python, HTML, CSS, Java, C#, Go).
• Bootstrap interview: At session start, the AI detects placeholders in project.config.md and asks 4 questions before the Engineering Interview: project name, author/company, license, and Git version control.
• File headers: Every new source file must include a copyright header adapted to the language. The dbv-specs-ops credit line is mandatory in all headers.
• LICENSE auto-generation: The AI generates the LICENSE file after the bootstrap interview based on the chosen license.
• Git integration: If the user accepts Git in bootstrap: git init, .gitignore generation (stack-aware), and first commit. On each /ship: Conventional Commits message, version tag, and push suggestion (not auto-executed).
• CHANGELOG.md auto-management: The AI accumulates entries in [Unreleased] during /build and /test, and publishes them under a new versioned section on each /ship.
• Semantic versioning on /ship: The AI presents a clear 4-option menu: Patch / Minor ✅ / Major / No change.
• Multiplatform startup scripts on /ship: Always generates start.cmd + stop.cmd (Windows) and start.sh + stop.sh (macOS/Linux), with automatic venv activation for Python projects.
• Python venv: Python projects always get a local virtual environment (venv/) created before installing dependencies in /build.

​

Changed

• Quick Start (EN + ES) completely rewritten with per-platform table and explicit first message.
• GEMINI.md, CLAUDE.md, .github/copilot-instructions.md, .windsurfrules: All updated with project.config.md in the startup reading list and two-step state detection (bootstrap check → specs check).
• MASTER_PROMPT.md: Bootstrap section, /build CHANGELOG rule, /test CHANGELOG rule, and full /ship protocol added.
• README file structure tables (EN + ES): project.config.md and CHANGELOG.md added.

​

Added

• Bilingual README (English / Spanish) with full Table of Contents and HTML anchors for navigation.
• ANTIGRAVITY.md: Dedicated setup file for Antigravity (VS Code · Google DeepMind): Planning Mode artifacts, Knowledge Items, and Context Snapshot instructions.
• Windsurf support: .windsurfrules file for automatic context loading in Windsurf.
• Visual Workflow diagram (Mermaid) added to README.
• docs/ADOPTION_PROMPT.md: Flow for adopting SDD in existing projects without specs.
• docs/README.md: Index of all files in the /docs folder.

​

Changed

• Core documentation files renamed to English for universal AI compatibility: ESPECIFICACIONES.md → SPECIFICATIONS.md, ARQUITECTURA.md → ARCHITECTURE.md.
• Artifact names standardized: task.md, implementation_plan.md, walkthrough.md — compatible with Antigravity’s native Planning Mode.
• GEMINI.md updated with Antigravity-specific behavior section (Planning Mode + Knowledge Items).
• Credit line for David Bueno Vallejo added consistently across all platform files.

​

Added

• docs/MASTER_PROMPT.md: The brain of the system — Senior Engineer rules, Spec→Plan→Build→Test→Simplify→Ship cycle, development standards, and boundaries.
• docs/SPECIFICATIONS.md: Template for project requirements: context, objectives, users, features, out-of-scope, risks, and open questions.
• docs/ARCHITECTURE.md: Template for technical decisions: stack, directory structure, key decisions, integrations, and MCP section.
• task.md: Logbook template with backlog, in-progress tracking, and Context Snapshot for session continuity.
• CLAUDE.md: Automatic context loading for Claude Code, Claude Desktop, and Cursor.
• GEMINI.md: Automatic context loading for Gemini CLI and Antigravity.
• .github/copilot-instructions.md: Automatic context loading for GitHub Copilot.
• project.config.md: Placeholder for project identity.
• README.md: Project documentation with origin, workflow, and usage instructions.
• LICENSE: MIT License.