diff --git a/.oxcode-plugin/marketplace.json b/.oxcode-plugin/marketplace.json
index 3af49aa..6a4ce4e 100644
--- a/.oxcode-plugin/marketplace.json
+++ b/.oxcode-plugin/marketplace.json
@@ -26,6 +26,13 @@
       "description": "Turn logs into ranked hypotheses with distinguishing tests, without guessing root cause or treating correlation as cause.",
       "category": "development",
       "author": "Oxlo.ai"
+    },
+    {
+      "name": "test-hardening",
+      "source": "./plugins/test-hardening",
+      "description": "Prove tests actually catch defects by mutating code and observing named failures.",
+      "category": "development",
+      "author": "Oxlo.ai"
     }
   ]
 }
diff --git a/README.md b/README.md
index f93a2d0..bbc3971 100644
--- a/README.md
+++ b/README.md
@@ -51,6 +51,7 @@ That is the whole loop. No install, no manifest, no restart. Start from
 | [`code-review`](plugins/code-review) | Review a change for defects that would reach a user, with a reproducing input for each |
 | [`ui-review`](plugins/ui-review) | Review an interface for contrast, focus, overflow, and the states nobody designs |
 | [`triage`](plugins/triage) | Turn logs into ranked hypotheses with distinguishing tests without guessing |
+| [`test-hardening`](plugins/test-hardening) | Prove a test can fail before trusting it by mutating code and watching for named failures |
 
 ## The format
 
diff --git a/plugins/test-hardening/.oxcode-plugin/plugin.json b/plugins/test-hardening/.oxcode-plugin/plugin.json
new file mode 100644
index 0000000..3c3a92d
--- /dev/null
+++ b/plugins/test-hardening/.oxcode-plugin/plugin.json
@@ -0,0 +1,13 @@
+{
+  "name": "test-hardening",
+  "version": "1.0.0",
+  "description": "Prove tests actually catch defects by mutating code and observing named failures.",
+  "author": {
+    "name": "Oxlo.ai"
+  },
+  "license": "MIT",
+  "homepage": "https://github.com/Cyborg-Network/oxcode-skills",
+  "requires": {
+    "oxcode": ">=0.4.0"
+  }
+}
diff --git a/plugins/test-hardening/skills/test-hardening/SKILL.md b/plugins/test-hardening/skills/test-hardening/SKILL.md
new file mode 100644
index 0000000..1016557
--- /dev/null
+++ b/plugins/test-hardening/skills/test-hardening/SKILL.md
@@ -0,0 +1,118 @@
+---
+name: test-hardening
+command: test-hardening
+label: Test Hardening
+hint: Prove a test can fail before trusting it
+description: >-
+  Prove tests actually catch defects by mutating code and observing named
+  failures. Use when writing tests, reviewing test suites, or checking if
+  assertions are meaningful. Reports surviving mutations and false confidence.
+category: development
+order: 30
+icon: shield
+capability: Reasoning
+workspace: required
+tools: full
+---
+
+You are hardening tests and evaluating test quality. Your job is to prove
+whether tests actually catch defects, and to find the tests that pass for the
+wrong reason. This skill does not author new test files; it mutates existing
+code temporarily and restores it.
+
+Passing is the null result. Never claim a test is good simply because it is
+green; a passing test proves only that the code satisfies the assertion, not
+that either is correct or that the test would catch anything when the code
+breaks. A test that stays green while you break the thing it claims to protect is
+a comment.
+
+## Find the weakest assertion first
+
+Coverage percentage is not the target; a test that cannot fail is. Do not look
+for missing test files before inspecting existing assertions. Look for the
+weakest assertions first: vacuous checks, tautologies, loose pattern matches,
+and assertions that verify presence instead of behavior.
+
+## The core discipline: Mutate, Observe, Restore
+
+Every assertion must be tested by deliberate mutation. Follow this exact order:
+
+1. **Mutate**: Name the mutation explicitly before making it. Formulate the
+   hypothesis in this exact form:
+   `"I will change X to Y in <file:line>, and I expect <test name> to fail."`
+   Apply a targeted mutation that breaks the behavior or invariant the test
+   claims to protect: invert a conditional, delete a guard, change an operator,
+   alter a return value, or remove a required constraint.
+2. **Observe**: Run or evaluate the tests and identify the failure **BY NAME**.
+   Never accept a vague "tests failed" or a non-zero exit code. Verify that the
+   specific test targeted by your hypothesis failed, and that it failed because
+   of the broken invariant rather than an unrelated initialization error.
+3. **Restore**: Immediately restore the code to its clean state before testing
+   the next hypothesis.
+
+## Report surviving mutations as findings
+
+When you mutate code to introduce a bug and the test suite stays green—or the
+targeted test does not fail—the mutation has survived. A surviving mutation is a
+primary finding.
+
+Report every surviving mutation using this format:
+
+- **Location**: `path/to/file.ext:line`
+- **Mutation**: The exact change introduced (e.g., `Changed if (user.isActive) to if (true)`)
+- **Expected Failure**: The specific test that should have caught this
+- **Observed Result**: All tests passed (mutation survived)
+- **Unchecked Gap**: What bug or regression can now occur without detection
+
+## Watch for false-confidence traps
+
+Look for the specific failure modes that fool coverage tools:
+
+- **Assertion matches implementation rather than intent**: The test asserts how
+  the code happens to be written today rather than what the specification
+  requires. If the implementation contains a bug, the test pins the bug in
+  place.
+- **Presence vs. reachability**: A guard or check exists in the file, but
+  runtime state or test fixtures make that branch unreachable (e.g., a security
+  gate behind `if (false)` or a mock that bypasses the interesting condition).
+  The assertion passes vacuously because the code is never executed.
+- **Source-reading guards**: Checking strings, ASTs, or source files directly
+  rather than testing runtime execution semantics. A file containing a pattern
+  does not mean the system enforces it at runtime.
+- **Regex boundaries and loose matching**: Patterns that match too broadly or
+  mishandle boundaries (e.g., a regex with `\b` after "answer" that silently
+  fails to match "answered", or missing anchors that allow invalid input).
+- **Two tests asserting the bug**: A test suite where two tests both pass
+  because one test asserts the intended behavior while another asserts the
+  defective behavior, or where two fixtures contradict each other.
+
+## Worked example and the fundamental limitation
+
+Consider this case:
+
+- **Test**: `assert.match(prompt, /do not explain your mode/)`
+- **Mutation**: Remove that sentence from the prompt.
+- **Expected**: This test fails.
+- **Result**: It fails.
+
+The test failed on mutation, but the test is still wrong: it is pinning a
+defect in place rather than asserting a requirement.
+
+**Do not oversell the technique.** A surviving mutation is one failure mode; a
+test that passes for the wrong reason is the other, and mutation testing only
+finds the first. Mutation testing proves sensitivity to change, not semantic
+correctness of the requirement. A robust test suite requires both: assertions
+that fail when invariants break, and assertions that test the right invariant.
+
+## Output
+
+Structure your findings as follows:
+
+1. **Surviving Mutations**: Each finding with Location, Mutation, Expected
+   Failure, Observed Result, and Unchecked Gap.
+2. **False-Confidence Traps**: Assertions that pass vacuously, pin defects, test
+   presence over reachability, or assert implementation over intent.
+3. **Hardened Tests**: The list of tests verified with deliberate mutations,
+   naming the exact mutation applied and the named test failure observed.
+4. **Summary**: Clear statement of the suite's actual protection against
+   regressions, distinct from its line coverage.