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.

ChatGPT (chat.openai.com) and Gemini Web (gemini.google.com) do not have access to your filesystem and do not auto-load any project files. This means the SDD workflow must be activated manually at the start of each conversation by pasting the contents of docs/MASTER_PROMPT.md directly into the chat. Once pasted, the AI operates under the same six-phase Spec-Driven Development lifecycle as any auto-loading platform.
For the full SDD experience — automatic context loading, file reads, Context Snapshots, and persistent memory across sessions — use an IDE-integrated platform such as Claude Code, Gemini CLI, Antigravity, Cursor, or Windsurf. Web interfaces are a useful fallback when IDE access is not available.

​
Why Manual Activation Is Required

IDE-integrated tools like Claude Code or Windsurf can read files from your project directory. Web AI assistants run in isolated browser sessions with no access to your local filesystem. There is no equivalent of CLAUDE.md or .windsurfrules that these interfaces can auto-detect. Every conversation starts with a completely blank context — no memory of previous sessions, no access to task.md or memory.md, and no awareness of your project’s specifications. Manual activation bridges this gap by injecting the workflow directives at the start of the conversation.

​
How to Activate Manually

​
Step 1 — Open docs/MASTER_PROMPT.md

Locate docs/MASTER_PROMPT.md in your project and open it in a text editor. This file is the brain of the SDD system — it defines the AI’s role, the six-phase workflow, the coding standards, and the trust boundary rules.

​
Step 2 — Paste the content as your first message

Copy the entire contents of docs/MASTER_PROMPT.md and paste it as the first message in a new ChatGPT or Gemini Web conversation. Then add the following after the pasted content: 
Review task.md and start the Engineering Interview for a new project.
Or, if you are resuming existing work: 
Here is the current task.md content:
[paste the contents of task.md]

Here is the current memory.md content:
[paste the contents of memory.md]

Resume from the last Context Snapshot in task.md.

​
Step 3 — Paste project context files manually

Since the AI cannot read files directly, you must supply any project context as text in the chat. For a new project, the Engineering Interview will generate the initial specs for you. For an existing project, paste the relevant files before making requests:
File to pasteWhen to include it
docs/SPECIFICATIONS.mdAlways — defines what you are building
docs/ARCHITECTURE.mdFor build, test, or refactor tasks
task.mdAlways — tells the AI where work left off
memory.mdFor any session longer than a single task
docs/DESIGN.mdFor UI/frontend tasks

​
Limitations Compared to IDE-Integrated Platforms

CapabilityIDE PlatformsChatGPT / Gemini Web
Auto-load activation file✅ Automatic❌ Manual paste required
Read project files directly✅ Full filesystem access❌ You must paste content
Write files to disk✅ Directly❌ Copy output manually
Persist context between sessions✅ Via task.md and memory.md❌ Each conversation is isolated
Context Snapshots✅ Written to task.md automatically⚠️ Must be copied manually
Knowledge Items (Antigravity)✅ Native❌ Not available

​
Handling Context Snapshots Manually

In IDE-integrated platforms, the AI writes a Context Snapshot to task.md at the end of each session. In a web conversation, you must do this yourself. At the end of a productive session, ask the AI: 
Write a Context Snapshot for task.md summarizing what was accomplished
and the exact next step to resume this work.
Copy the AI’s output and paste it into your local task.md file. In the next session, include this snapshot when you paste task.md content so the AI can resume from the correct point.

​
Best Practices for Long Sessions

Web AI conversations have context length limits. As a session grows longer, the AI may lose track of instructions or earlier decisions. Apply these practices to maintain quality:
  • Start every session fresh: Open a new conversation and re-paste MASTER_PROMPT.md. Do not rely on the AI to remember rules from a previous conversation.
  • Include memory.md and task.md at the start of each session: These two files together give the AI the qualitative history and current state of the project. Paste both at the same time as MASTER_PROMPT.md.
  • Keep sessions focused on one phase: Complete a single phase (e.g., /spec or /build) per conversation to minimize context drift.
  • Copy all outputs to disk immediately: When the AI produces code, updated specs, or a Context Snapshot, copy it to the corresponding file in your project before closing the browser tab. Web conversations do not persist.
  • Re-inject context if the AI starts deviating: If the AI stops following the SDD workflow mid-conversation, paste the relevant section of MASTER_PROMPT.md again as a reminder.
Web AI sessions are ephemeral. If you close a ChatGPT or Gemini Web conversation without saving the AI’s outputs to your local project files, that work is permanently lost. Always copy code, specs, and context snapshots to disk before ending a session.

Build docs developers (and LLMs) love

Get started for freeTalk to us