/eiffel.harden - Phase 6: Adversarial Testing

Purpose: Break the implementation with adversarial inputs, stress tests, and edge cases.

Usage

code

/eiffel.harden <project-path>

Project Scoping

code

<project-path>/
├── .eiffel-workflow/
│   ├── prompts/
│   │   └── phase6-adversarial-review.md
│   └── evidence/
│       └── phase6-tests.txt
├── test/*.e (adversarial tests added here)

Prerequisites

•Phase 5 complete: all tests pass

Workflow

Step 0: MML Contract Hardening

Before adversarial tests, strengthen contracts with Mathematical Model Library (MML).

Add simple_mml dependency to ECF:

xml

<library name="simple_mml" location="$SIMPLE_EIFFEL\simple_mml\simple_mml.ecf"/>

Add MML model queries for collections:

eiffel

feature -- Model Queries

    items_model: MML_MAP [K, V]
            -- Mathematical model of stored items.
        do
            create Result.make_empty
            across internal_table as ic loop
                Result := Result.extended (@ic.key, @ic.item)
            end
        end

Strengthen postconditions with frame conditions:

eiffel

ensure
    -- What changed
    key_added: items_model.has (a_key)
    value_set: items_model [a_key] = a_value

    -- What did NOT change (frame condition)
    others_unchanged: items_model.removed (a_key) |=| old items_model.removed (a_key)

MML Checklist:

• Collection attributes have model queries (MML_SET, MML_MAP, MML_SEQUENCE)
• Postconditions use |=| for equality comparisons
• Frame conditions specify what did NOT change
• Old expressions capture pre-state: old items_model

Step 1: Generate Adversarial Prompt

Create <project-path>/.eiffel-workflow/prompts/phase6-adversarial-review.md:

markdown

# Adversarial Test Suggestions

## Contracts
[PASTE src/*.e]

## Current Tests
[PASTE test/*.e]

## Suggest Tests For:
1. Boundary values (0, 1, MAX_INT, empty strings)
2. Capacity limits (full, zero capacity)
3. SCOOP concurrent access
4. State transitions in unexpected order
5. Resource exhaustion

Step 2: Human Submits to AI (Optional)

User may submit prompt to external AI for adversarial suggestions.

Step 3: Generate Adversarial Tests

Add tests like:

•test_with_zero_capacity
•test_same_key_repeatedly
•test_stress_many_operations
•test_concurrent_access_scoop

Step 4: Compile and Run All Tests

CRITICAL: You MUST cd to the project directory before compiling. The EIFGENs folder is created in the current working directory, not where the ECF file is located. Compiling from the wrong directory pollutes other folders with build artifacts.

bash

cd <project-path> && /d/prod/ec.sh -batch -config <project>.ecf -target <project>_tests -c_compile
./EIFGENs/<project>_tests/W_code/<project>.exe

ZERO WARNINGS POLICY: If the compiler reports obsolete call warnings or any other warnings, FIX THEM IMMEDIATELY. Do not note them and move on. Do not defer them. Fix them right now before proceeding.

ALL TESTS MUST PASS.

Step 4b: SCOOP Consumer Integration Gate (MANDATORY)

Why: This catches integration issues that don't appear in standalone compilation. A library with concurrency=scoop may compile fine but cause VUAR(2) errors in consumer libraries.

Create mock consumer project (temporary, can delete after test):

bash

mkdir -p <project-path>/test_consumer

Create <project-path>/test_consumer/consumer_test.ecf:

xml

<?xml version="1.0" encoding="ISO-8859-1"?>
<system name="consumer_test" uuid="[generate-new-uuid]">
    <target name="consumer_test">
        <root class="CONSUMER_TEST" feature="make"/>
        <capability>
            <concurrency support="scoop"/>
            <void_safety support="all"/>
        </capability>
        <library name="base" location="$ISE_LIBRARY\library\base\base.ecf"/>
        <library name="<project>" location="..\<project>.ecf"/>
        <cluster name="src" location="."/>
    </target>
</system>

Create <project-path>/test_consumer/consumer_test.e:

eiffel

class CONSUMER_TEST
create make
feature
    make
        local
            l_obj: <MAIN_LIBRARY_CLASS>
        do
            create l_obj.make
            -- Use library types in SCOOP context
        end
end

Compile consumer:

bash

cd <project-path>/test_consumer && /d/prod/ec.sh -batch -config consumer_test.ecf -target consumer_test -c_compile

If VUAR(2) errors: Fix generic constraints in the library (add separate keyword).

Cleanup:

bash

rm -rf <project-path>/test_consumer

Step 5: Save Evidence

Save to <project-path>/.eiffel-workflow/evidence/phase6-tests.txt:

code

# Phase 6 Hardening Evidence
# Project: <project-path>
# Date: [timestamp]

## MML Hardening
MML dependency added: [yes/no]
Model queries added: [count]
Frame conditions added: [count]

## SCOOP Consumer Integration
Consumer test created: [yes/no]
Consumer compilation: [PASS/FAIL]
VUAR(2) errors fixed: [count or N/A]

## Adversarial Testing
Adversarial tests added: [count]
Stress tests added: [count]

[Paste test output]

# Status: PASS / FAIL

Context Management (RLM Pattern)

This skill focuses ONLY on: <project-path>

DO NOT:

•Read files outside this project directory
•Load entire ecosystem or multiple libraries into context
•Keep large file contents in working memory

DO:

•Use Task tool with Explore agent for ecosystem questions
•Ask targeted questions when adversarial tests need to understand attack vectors from similar libraries
•Release context after getting answers

Example - Ecosystem Query:

code

Need ideas for stress testing patterns? Don't read all test files in ecosystem.
Instead: Task(subagent_type=Explore) → "What adversarial test patterns exist in simple_container tests?"

The sub-agent searches, summarizes, and returns ONLY what you need. Your context stays focused on this project.

Completion

code

Phase 6 COMPLETE: Hardening passed.
Project: <project-path>

All tests: PASS

Next: Run /eiffel.ship <project-path> for final checklist.