Hardening Skill
What It Does
Hardening sharpens what already exists — tests, ISCs, and code — without adding new functionality. Five workflows target test-surface and code-surface robustness: property-based testing via fast-check, mutation testing via Stryker, CRAP-complexity scoring, DRY duplication detection, and acceptance-test mutation that perturbs ISC text to catch fluff.
The Problem
Example-based tests check a handful of inputs the author thought of, so the bugs that survive are the inputs the author didn't think of. A test suite can be green and still be weak. ISCs can read as crisp pass/fail and still be fluff that anything would satisfy. You can't see any of this by reading the tests — you need to test the tests. This skill mechanizes that: property tests express the universal claim and shrink failures to the minimal counterexample, mutation testing proves the suite catches injected bugs, and acceptance-test mutation proves each ISC actually constrains the work.
How It Works
Hardening techniques sharpen what already exists — tests, ISCs, code. They don't add new functionality. Five workflows, all targeting test-surface and code-surface robustness.
Workflow Routing
| Workflow | Trigger | File |
|---|
| PropertyTest | property test, property based testing, PBT, fast-check, universal quantified claim, shrink counterexample — pure functions, parsers, serializers, data transforms, invariants | Workflows/PropertyTest.md |
| MutationTest | mutation test, mutation testing, Stryker, test the tests | — planned, not yet built (stub; see Status) |
| CrapAnalysis | CRAP score, CRAP analysis, risky undertested code | — planned, not yet built (stub; see Status) |
| DryAnalysis | DRY scan, jscpd, duplication rot | — planned, not yet built (stub; see Status) |
| AcceptanceTestMutation | acceptance test mutation, sharpen ISCs, detect fluff ISCs | — planned, not yet built (stub; see Status) |
Doctrine
These workflows operate against the existing test surface. PropertyTest doesn't replace bun-test examples — properties express the universal claim, examples are sampled instances of it. MutationTest doesn't replace test authoring — it grades existing tests' robustness. CRAP and DRY don't add tests — they prioritize where to add them. Acceptance-test mutation doesn't add ISCs — it mechanizes the Fluff vs Load-bearing distinction from IsaFormat.md.
All five satisfy the same intent: strengthen the test of the test, the test of the ISC, the test of the code. The unifying frame is meta-test — testing the things that test the system.
Status
| Workflow | State | Blocker |
|---|
| PropertyTest | Fully scaffolded (v1.0) | None — ready to use |
| MutationTest | Stub | Stryker integration; deferred to v6.11.0 |
| CrapAnalysis | Stub | AST walker (oxc or bun build --print-ir) |
| DryAnalysis | Stub | jscpd wrapper |
| AcceptanceTestMutation | Stub | ISC text perturbation generator |
Integration Points
- Testing Doctrine Rule #11 — names fast-check as the property-testing primitive (
LIFEOS/DOCUMENTATION/Testing/TestingDoctrine.md).
- ISA format — new
bun-property ISC type with property | generator | runs columns (LIFEOS/DOCUMENTATION/Isa/IsaFormat.md § ISC Type Vocabulary).
- Algorithm v6.10.0 candidate — VERIFY-phase Hardening gate at E4/E5;
PropertyAudit as capability #20 (LIFEOS/ALGORITHM/v6.10.0.md).
- System prompt — Verification Is the Mechanism section at top (
LIFEOS/LIFEOS_SYSTEM_PROMPT.md).
- BitterPillEngineering skill —
AcceptanceTestMutation is the mechanized form of BPE's "would a smarter model render this rule unnecessary" applied to ISCs.
Gotchas
- fast-check shrinking is deterministic only when the seed is captured. Always pin the seed in a comment when a property fails:
// fc seed: 0xdeadbeef.
numRuns: 1000 is the default budget. Increase to 10000 for invariant-critical properties; lower to 100 for slow generators (custom record types with large constraints).- Property tests spuriously fail with poorly-constrained generators. Constrain integer ranges, string lengths, and record nesting depth explicitly.
- Properties on impure functions (functions that touch disk, network, clock) are wrong — fast-check generates random inputs but the property must be deterministic given the input. Either mock side effects or extract the pure core.
- Generator over-constraint hides bugs (
fc.integer({min: 0, max: 100}) for code that handles all integers). Constrain only what the function actually demands.
- Generator under-constraint produces invalid inputs (
fc.string() when the function only accepts ASCII). The property must hold across the function's actual valid input domain, no wider.
Examples
- "Property-test the slug parser" → PropertyTest: express round-trip/idempotence claims as fast-check properties, run 1000 random inputs, pin the seed on any failure.
- "What bugs are my example tests missing?" → PropertyTest candidate detection: scan for pure functions (parsers, serializers, transforms) with only example coverage.
- "Mutation-test the hooks suite" → MutationTest is a stub (see Status) — not runnable yet; don't promise kill-rate numbers.