majiayu000 PHANTOM GUARD — FORTRESS 2.0 MASTER PROTOCOL
Classification: MILITARY-GRADE DEVELOPMENT PROTOCOL Project: Phantom Guard - Slopsquatting Detection Framework: FORTRESS 2.0 Violation Response: STOP. FIX. CONTINUE.
MISSION STATEMENT
Build a slopsquatting detection library that protects developers from AI-hallucinated malicious packages.
- Target: <5% false positive rate
- Speed: <200ms per package
- Scope: PyPI, npm, crates.io
GATE SYSTEM — MANDATORY PROGRESSION
┌─────────────────────────────────────────────────────────────────────┐
│ FORTRESS 2.0 GATE SYSTEM │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ GATE 0: PROBLEM DEFINITION │
│ Status: ✅ COMPLETE (Research Fortress v7) │
│ Output: PROJECT_FOUNDATION.md │
│ │
│ GATE 1: ARCHITECTURE │
│ Command: /architect │
│ Status: 🟡 PENDING │
│ Output: docs/architecture/ARCHITECTURE.md │
│ ⛔ STOP: Every decision needs SPEC_ID, HOSTILE_ARCHITECT review │
│ │
│ GATE 2: SPECIFICATION │
│ Command: /spec │
│ Status: ⬜ BLOCKED (needs Gate 1) │
│ Output: docs/specification/SPECIFICATION.md │
│ ⛔ STOP: Every invariant numbered, every edge case documented │
│ │
│ GATE 3: TEST DESIGN │
│ Command: /test │
│ Status: ⬜ BLOCKED (needs Gate 2) │
│ Output: tests/ stubs, TEST_MATRIX.md │
│ ⛔ STOP: Tests exist BEFORE code, coverage targets defined │
│ │
│ GATE 4: PLANNING │
│ Command: /roadmap │
│ Status: ⬜ BLOCKED (needs Gate 3) │
│ Output: docs/planning/ROADMAP.md with traced tasks │
│ ⛔ STOP: Every task < 8 hours, every task traces to SPEC_ID │
│ │
│ GATE 5: VALIDATION │
│ Command: /hostile-review │
│ Status: ⬜ BLOCKED (needs implementation) │
│ Output: VALIDATION_REPORT.md │
│ ⛔ STOP: HOSTILE_VALIDATOR has VETO power │
│ │
│ GATE 6: RELEASE │
│ Command: /release │
│ Status: ⬜ BLOCKED (needs Gate 5) │
│ Output: Release artifacts, CHANGELOG.md │
│ ⛔ STOP: RELEASE_GUARDIAN final sign-off │
│ │
└─────────────────────────────────────────────────────────────────────┘
COMMAND INDEX
Gate Commands (Sequential - Cannot Skip)
| Command | Gate | Purpose | Can Skip? |
|---|---|---|---|
/architect |
1 | System architecture design | NO |
/spec |
2 | Specification, invariants | NO |
/test |
3 | Test design before code | NO |
/roadmap |
4 | Task breakdown | NO |
/hostile-review |
5 | Validation (VETO POWER) | NO |
/release |
6 | Release preparation | NO |
Implementation Commands
| Command | Purpose |
|---|---|
/implement |
Guided TDD implementation |
/competitive-watch |
Scan for competition (weekly) |
/validate-technical |
Validate APIs still work (monthly) |
ABSOLUTE RULES — NO EXCEPTIONS
Rule 1: Gates Cannot Be Skipped
❌ FORBIDDEN: "Let's skip architecture and start coding"
❌ FORBIDDEN: "We'll add tests later"
❌ FORBIDDEN: "This is simple, no spec needed"
✅ REQUIRED: Complete Gate N before starting Gate N+1
✅ REQUIRED: Gate outputs must exist before proceeding
✅ REQUIRED: If blocked, fix the blocker first
Rule 2: HOSTILE_VALIDATOR Has Veto Power
If HOSTILE_VALIDATOR says NO_GO:
1. STOP all work
2. Address every issue raised
3. Re-run validation
4. Only proceed after GO verdict
NO EXCEPTIONS. NO SHORTCUTS. NO "WE'LL FIX IT LATER."
Rule 3: TDD Is Mandatory
Before writing ANY production code:
1. Test stub MUST exist
2. Test MUST fail when run
3. Write ONLY enough code to pass test
4. Refactor if needed
5. THEN commit
Rule 4: Trace Everything
Every function: # IMPLEMENTS: S001
Every test: # SPEC: S001, T001.1
Every task: TRACES: S001, INV001
Orphan code (no trace) = BUILD FAILURE
Rule 5: Performance Budget Is Sacred
| Operation | Budget | Violation = |
|-----------|--------|-------------|
| Single package (cached) | <10ms | BLOCK MERGE |
| Single package (uncached) | <200ms | BLOCK MERGE |
| 50 packages | <5s | BLOCK MERGE |
No exceptions. No "we'll optimize later."
SESSION START PROTOCOL
Every development session MUST begin with:
1. /master # Load this context
2. Check .fortress/FORTRESS.md # Know gate status
3. Proceed with current gate # Continue work
FILE STRUCTURE
phantom-guard/
├── .fortress/ # FORTRESS 2.0 config
│ ├── FORTRESS.md # Framework status
│ └── gates/ # Gate completion records
├── .claude/skills/ # Command definitions
│ ├── master/ # This command
│ ├── architect/ # Gate 1
│ ├── spec/ # Gate 2
│ ├── test/ # Gate 3
│ ├── roadmap/ # Gate 4
│ ├── hostile-review/ # Gate 5
│ ├── release/ # Gate 6
│ ├── implement/ # TDD workflow
│ ├── competitive-watch/ # Competition scan
│ └── validate-technical/ # API validation
├── docs/
│ ├── frameworks/ # FORTRESS 2.0 docs
│ └── research/ # Technical research
└── PROJECT_FOUNDATION.md # SOURCE OF TRUTH
CURRENT STATUS
Active Gate: Gate 1 (Architecture)
Next Action: Run /architect to begin architecture design
Blockers: None
SOURCE OF TRUTH
- PROJECT_FOUNDATION.md - Original research
- .fortress/FORTRESS.md - Framework configuration
- docs/frameworks/FORTRESS_2.0_FRAMEWORK.md - Full framework docs
FORTRESS 2.0 — Because bugs escape reviews, but they don't escape gates.