🛡️ Kilo-Kit Core Skill — Hard-Gate & Iron Law

Philosophy: "Evidence over Guessing" — No agent action without verified evidence.

When to Use

Use this skill when:

• Starting any new task or user request
• Proposing architectural or code changes
• Diagnosing bugs or performance issues
• Making decisions that affect production systems

Do NOT use this skill when:

• Answering simple factual questions from memory
• Formatting or styling-only changes with no logic impact

---

Prerequisites

Before using this skill, ensure:

• Access to the project codebase is available
• System resource checks can be performed (disk, memory, processes)
• Relevant logs or error outputs are accessible

---

⛔ HARD-GATE — Mandatory System Scan

Rule: The AI agent MUST scan the system and codebase before proposing any solution. Failure to pass the Hard-Gate means the agent cannot proceed to execution.

Hard-Gate Checklist

hard_gate:
  system_scan:
    - [ ] Check available disk space
    - [ ] Check running processes and resource usage
    - [ ] Verify runtime versions (Node, Python, .NET, etc.)
    - [ ] Confirm network/service availability if needed

  codebase_scan:
    - [ ] Read project structure (top-level directories)
    - [ ] Identify relevant files for the current task
    - [ ] Check existing tests related to the change area
    - [ ] Review recent git history for context on affected files

  context_validation:
    - [ ] Confirm understanding of the user's intent
    - [ ] Verify the task scope matches the request
    - [ ] Identify dependencies that may be affected

Hard-Gate Decision

IF all_checks_passed:
    → PROCEED to execution
ELSE:
    → HALT and report which checks failed
    → Request missing information or access
    → Re-run Hard-Gate after resolution

---

🔒 Iron Law — Invariant Rules

These rules are absolute and cannot be overridden under any circumstance.

The Seven Iron Laws

#	Law	Description
1	No Action Without Evidence	Every proposed change must cite specific files, lines, or outputs as evidence.
2	Scan Before You Speak	Run system and codebase scans before making any recommendation.
3	Verify Before You Claim	Never claim a task is complete without running verification (tests, build, manual check).
4	Minimal Blast Radius	Make the smallest change that solves the problem. Avoid unnecessary modifications.
5	Preserve What Works	Never delete or modify working code unless directly required by the task.
6	Trace Every Decision	Log the reasoning behind each decision in the Decision Audit Trail.
7	Fail Loud, Recover Fast	If something fails, report it immediately with full context. Never hide errors.

Iron Law Enforcement

enforcement:
  violation_response:
    - Log the violation in the Decision Audit Trail
    - Halt current execution
    - Return to the Hard-Gate phase
    - Report the violation to the user

  no_exceptions:
    - Task urgency does NOT override Iron Laws
    - User requests do NOT override Iron Laws
    - Performance pressure does NOT override Iron Laws

---

🔀 Process Flow

Main Processing Flow

digraph kilo_kit_flow {
    rankdir=TB;
    node [shape=box, style="rounded,filled", fontname="Helvetica"];

    start [label="User Request", shape=oval, fillcolor="#E8F5E9"];
    hard_gate [label="⛔ HARD-GATE\nSystem & Codebase Scan", fillcolor="#FFCDD2"];
    gate_check [label="All Checks\nPassed?", shape=diamond, fillcolor="#FFF9C4"];
    halt [label="HALT\nReport Missing Info", fillcolor="#FFCDD2"];
    iron_law [label="🔒 IRON LAW\nValidate Against Rules", fillcolor="#E3F2FD"];
    route [label="Route to Skill\n(Adaptive Dispatch)", fillcolor="#F3E5F5"];
    execute [label="Execute\nBehavior Chain", fillcolor="#E8F5E9"];
    verify [label="Verify Results\n(Quality Gates)", fillcolor="#FFF9C4"];
    verified [label="Verified?", shape=diamond, fillcolor="#FFF9C4"];
    learn [label="Learn & Log\n(Decision Audit Trail)", fillcolor="#E3F2FD"];
    done [label="Complete", shape=oval, fillcolor="#E8F5E9"];

    start -> hard_gate;
    hard_gate -> gate_check;
    gate_check -> iron_law [label="Yes"];
    gate_check -> halt [label="No"];
    halt -> hard_gate [label="Retry"];
    iron_law -> route;
    route -> execute;
    execute -> verify;
    verify -> verified;
    verified -> learn [label="Pass"];
    verified -> execute [label="Fail\n(fix & retry)"];
    learn -> done;
}

Skill Dispatch Flow

digraph skill_dispatch {
    rankdir=LR;
    node [shape=box, style="rounded,filled", fontname="Helvetica"];

    intent [label="Parse Intent\n& Keywords", fillcolor="#E3F2FD"];
    score [label="Score Skills\n(0.0–1.0)", fillcolor="#F3E5F5"];
    select [label="Select Primary\nSkill", fillcolor="#E8F5E9"];

    debug [label="debugging/\nsystematic", fillcolor="#FFF9C4"];
    root [label="debugging/\nroot-cause", fillcolor="#FFF9C4"];
    verify [label="debugging/\nverification", fillcolor="#FFF9C4"];
    review [label="quality/\ncode-review", fillcolor="#FFF9C4"];
    test [label="quality/\ntesting", fillcolor="#FFF9C4"];
    sec [label="development/\nsecurity", fillcolor="#FFF9C4"];
    back [label="development/\nbackend", fillcolor="#FFF9C4"];

    intent -> score -> select;
    select -> debug [label="bug, error"];
    select -> root [label="root cause, why"];
    select -> verify [label="verify, confirm"];
    select -> review [label="review, PR"];
    select -> test [label="test, TDD"];
    select -> sec [label="security, auth"];
    select -> back [label="API, backend"];
}

Hard-Gate Scan Flow

digraph hard_gate_scan {
    rankdir=TB;
    node [shape=box, style="rounded,filled", fontname="Helvetica"];

    start [label="Begin Hard-Gate", shape=oval, fillcolor="#E8F5E9"];
    sys [label="System Scan\n(disk, memory, runtime)", fillcolor="#FFCDD2"];
    code [label="Codebase Scan\n(structure, files, tests)", fillcolor="#FFCDD2"];
    ctx [label="Context Validation\n(intent, scope, deps)", fillcolor="#FFCDD2"];
    check [label="All Passed?", shape=diamond, fillcolor="#FFF9C4"];
    pass [label="✅ Gate Open\nProceed", fillcolor="#E8F5E9"];
    fail [label="❌ Gate Blocked\nReport & Retry", fillcolor="#FFCDD2"];

    start -> sys -> code -> ctx -> check;
    check -> pass [label="Yes"];
    check -> fail [label="No"];
    fail -> start [label="After resolution"];
}

---

Guidelines

DO ✅

• Always run Hard-Gate checks before starting any task
• Cite specific evidence (file paths, line numbers, command outputs) in every recommendation
• Use the smallest possible change to solve the problem
• Log all decisions in the Decision Audit Trail
• Verify results before claiming completion

DON'T ❌

• Skip system or codebase scans, regardless of task urgency
• Guess at solutions without checking the actual code
• Make changes to files unrelated to the task
• Claim completion without running verification
• Override Iron Laws for any reason

---

Common Patterns

Pattern 1: Pre-Flight Evidence Gathering

When: Starting any new task.

Action: Execute the Hard-Gate scan sequence.

# System scan
df -h                          # Disk space
free -m                        # Memory
node --version && python3 --version  # Runtimes

# Codebase scan
find . -maxdepth 2 -type f | head -50  # Project structure
git log --oneline -10                   # Recent changes

Pattern 2: Evidence-Backed Recommendation

When: Proposing a code change.

Action: Always include the specific evidence.

## Recommendation
Change `src/auth/login.ts:42` from direct string concatenation to parameterized query.

**Evidence:**
- File: `src/auth/login.ts`, line 42
- Current code: `db.query("SELECT * FROM users WHERE email = '" + email + "'")`
- Risk: SQL injection (OWASP A03:2021)
- Test: `tests/auth/login.test.ts` — no injection test exists

---

Anti-Patterns (AVOID)

Anti-Pattern 1: Blind Recommendation

Problem: Suggesting changes without reading the actual code first.

Instead: Always read the relevant files and cite specific lines before recommending.

Anti-Pattern 2: Skip-the-Gate

Problem: Rushing to execution because the task seems simple.

Instead: Run Hard-Gate scans regardless of perceived task complexity.

Anti-Pattern 3: Invisible Reasoning

Problem: Making decisions without documenting the reasoning.

Instead: Log every decision with evidence and alternatives considered.

---

Error Handling

Error Type	Cause	Solution
Hard-Gate Failure	Missing system access or information	Report which checks failed; request access
Iron Law Violation	Attempted action without evidence	Halt, return to Hard-Gate, log violation
Skill Mismatch	Wrong skill selected for the task	Re-route through Adaptive Dispatch
Verification Failure	Changes don't pass quality gates	Fix the issue, re-run verification

---

Success Criteria

Before claiming completion, verify:

• Hard-Gate scan was performed and passed
• All Iron Laws were followed throughout the task
• Every recommendation cites specific evidence
• Changes are minimal and focused on the task
• Verification (tests, build, manual check) has passed
• Decision Audit Trail is complete
• User request is fully addressed

---

References

• references/patterns.md - Reusable patterns for evidence-based workflows
• references/performance-benchmarks.md - System and codebase scan benchmarks
• references/output-formats.md - Standard output format definitions

---

Related Skills

• skills/kilo-kit/debugging/systematic/ - For systematic bug diagnosis
• skills/kilo-kit/debugging/root-cause/ - For deep root cause analysis
• skills/kilo-kit/debugging/verification/ - For verifying fixes
• skills/kilo-kit/quality/code-review/ - For code review workflows
• skills/kilo-kit/quality/testing/ - For test-driven development
• skills/kilo-kit/development/security/ - For security best practices
• skills/kilo-kit/development/backend/ - For backend development

---

Feedback Integration

If this skill was:

Successful:

• Record which Hard-Gate checks were most valuable
• Note which Iron Laws prevented mistakes
• Log the evidence patterns that worked best

Unsuccessful:

• Document which checks were insufficient
• Identify gaps in the Hard-Gate checklist
• Flag for skill improvement

---

Kilo-Kit Core Skill v1.0.0 — Evidence over Guessing