The .wrf.md file is a single-file protocol for AI coding tasks. It carries the step roadmap, active step, command ledger, completed checkpoints with evidence, strategic notes, scope adjustments, branch tasks, and human Contribution notes. Any AI tool or model that can read and write Markdown can pause, switch, and continue the same work without losing context.
WRF Deck is a local browser-based visual workspace that interoperates with the protocol, not another AI editor. It reads the file, manages step transitions, generates next-step and branch task commands, enforces Sentry guardrails, and isolates human instructions from AI execution checkpoints. Copy a command, paste it into Claude Code / GPT / Copilot / Gemini / a script, and the tool executes the step and writes progress back. Your AI tools handle the actual work; Deck is a parallel state layer for the file.
Use the handoff buttons in WRF Deck to generate a structured context snapshot, or share the full .wrf.md file directly — both resume instantly at the exact same step. You own the file. Email it, commit it, or hand it to a teammate.
Sentry runs a two-layer scan: it checks the command ledger for declared dangerous commands, and also scans the full .wrf.md text for raw dangerous CLI patterns — so even if the AI skips the ledger, the command is still caught. When a match is found, the AI must stop and wait for your confirmation. You can lock the incident with one click to generate a SHA-256 hash + timestamp proof. It is a behavioral declaration layer, not a kernel-level sandbox — it won't intercept a process that bypasses the .wrf.md file entirely.
Repo has the live workspace, demo, and a starter template:
quanliangdaima
/
1G10K-wrf-protocol
WRF protocol: AI workflow security infrastructure with dangerous command sentry alerts, context handoff, multi-task pause/resume, and human-AI contribution evidence.
WRF DECK
Sentry, multi-AI handoff, and human-AI contribution evidence
English · 简体中文 · Français · Español · Português (Brasil) · 日本語 · Deutsch · हिन्दी
A local-first, single-file workspace for AI workflows — no install required.
Reference implementation of the open WRF protocol — one file, any AI tool, zero lock-in.
What problem does it solve?
- Dangerous commands. Executing without warning, forcing you to manually capture evidence under stress.
- Pause-and-resume cost. You cannot step away freely. Every meeting or break forces you to spend time and tokens rebuilding context.
- Context drift. Context window limits that cause drift, forgotten constraints, or confident hallucinations.
- Tool-switching overhead. Repeating goals, constraints, and progress every time you switch tools.
- Multi-task switching cost. Running multiple AI tasks at once makes continuous working state hard to manage.
- Team handoff overhead. Handing off work requires detailed explanations; remote voice or…
I'm around to answer questions and would appreciate feedback on the protocol format and the WRF Deck workspace.
Top comments (0)