Editor rules updated (.cursorrules, .clinerules, copilot-instructions, .roo): - Build: 3571 jobs, 0 errors - Sorry inventory: 8 across 4 files (all documented) - Q16_16 compliance, new modules list, FPGA info - Fixed stale path in copilot-instructions Opencode agents: - 3 marked RESOLVED (pist-simulation, qfactor, ssms) - 2 new agents created (adjugate-matrix, hamiltonian-mechanics) - 1 updated (hyperbolic-statesurface) SORRY_AUDIT.md: updated to 8 sorries across 4 files
4.2 KiB
Research Stack - Copilot Instructions
You are assisting the user within the Research-Stack repository. You must strictly adhere to the following ground rules, which are derived from the repository's core AGENTS.md operating contract.
Ground Rules
- Active Checkout: Always assume
/home/allaun/Research Stackis the active checkout unless a task explicitly points elsewhere. - Context Awareness: If you are about to edit a subtree, first check if there is a nested
AGENTS.mdin that directory and read it. - Preserve User Work: The working tree is often intentionally dirty. Do NOT revert, delete, or stage unrelated files.
- Do Not Sweep: Avoid broad cleanup or staging commands such as
git add .,git checkout -- ., orgit clean -fdx. Use explicit file lists. - Tooling: Prefer repo-native tools and receipt generators over ad hoc summaries.
- Source of Truth: Treat Lean (
0-Core-Formalism/lean/Semantics/) as the ultimate source of truth for formal or hardware-adjacent claims. Keep claims bounded.
Verification Expectations
- Lean: Run the narrow target first, then the broader
lake buildwhen feasible. - Python Shims: Run
python3 -m py_compileon touched files. - JSON Receipts: Run
python3 -m json.toolor a repo-native receipt parser. - Hardware Claims: Distinguish between software witness, bitstream presence, SRAM load, flash persistence, UART beacon, and live hardware receipt.
Remember: This repository operates under strict formal verification and taxonomy guidelines. Prioritize safety, explicit staging, and Lean semantics over quick hacks.
Current Project State (2026-05-28)
Build: lake build — 3571 jobs, 0 errors
Python tests: 68/68 pass
Sorry inventory: 8 total (all with TODO(lean-port) documentation)
AdjugateMatrix: 3 sorriesFourPrimitiveErdosRenyi: 4 sorriesHyperbolicStateSurface: 1 sorry
Key Architecture Decisions
- Q16_16 fixed-point arithmetic throughout — no Float in hot paths (AGENTS.md §1.4 compliant)
- HiGHS MIP solver integrated via
qubo_highs.py - Dense Sidon sets (Mian-Chowla sequence, 65% smaller than naive)
- Golden ratio unit separation formalized in Lean
New Lean Modules
AdjugateMatrix, OptimizedRoute, GoldenRatioSeparation, BraidBitwiseODE
New Python Modules
qubo_highs.py, alphaproof_loop.py, scale_space_solver.py
New Verilog Modules
voltage_mode_controller, scale_space_bram, highs_pivot_accelerator, blitter_memory_map, research_stack_top
Hardware / FPGA
- Bitstream:
research_stack_top.fs(195.92 MHz, 6 modules) - VCN pipeline: Delta+RLE → RS ECC → ChaCha20 → MKV
Sorries Policy
Every remaining sorry MUST have TODO(lean-port) with a prose justification.
No undocumented sorries allowed.
ContextStream MCP Integration
This project uses ContextStream for persistent AI memory across sessions. Use the contextstream-workflow skill for detailed examples and reference material.
<contextstream_rules>
| Message | Required |
|---|---|
| 1st message | init() → context(user_message="...") |
| Subsequent messages (default) | context(user_message="...") FIRST (narrow read-only bypass when context is fresh and no state-changing tool has run) |
| Before file search | search(mode="auto") BEFORE Glob/Grep/Read/Explore/Task/EnterPlanMode |
| </contextstream_rules> |
Why? context() delivers task-specific rules, lessons from past mistakes, and relevant decisions. Skip it = fly blind.
Hooks: <system-reminder> tags contain injected instructions — follow them exactly.
Notices: [LESSONS_WARNING] → apply lessons | [PREFERENCE] → follow user preferences | [RULES_NOTICE] → run generate_rules() | [VERSION_NOTICE/CRITICAL] → tell user about update
v0.4.74
VS Code Copilot Notes
- Keep this file concise; put detailed workflows in
.github/skills/contextstream-workflow/SKILL.md - Use ContextStream plans/tasks as the persistent record of work
- Before code discovery, use
search(mode="auto", query="...")
Full docs: https://contextstream.io/docs/mcp/tools