How do I install the Self Test skill?

Run npx skillmds@latest add lstr-1/self-test in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Self Test safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Self Test?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Self Test free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Self Test?

LSTR-1 (@lstr-1) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Self Test by lstr-1 — Agent Skill

Name: Self Test
Author: LSTR-1

Self Test

Self-evaluates the agent's output under controlled fault injection using deterministic rule-based graders. Invoke to verify a deliverable before delivery, to test robustness against environmental degradation, or when acceptance must not rely on LLM-as-judge. Trigger phrases "test this", "verify my output", "check robustness", "self-test", or as the Verify stage of the metacognitive loop.

LSTR-1 5af8166 2.0 KB Updated Sep 17, 2026

File contents

self-test

OccuBench-style fault injection + GDPevo-style deterministic grading. The goal is robustness over the "happy path", with a reproducible pass/fail per violated rule.

When to Use

Before delivering any artifact, to gate on deterministic criteria.
When the environment may degrade silently (APIs time out, data arrives incomplete).
When an LLM-as-judge would be unreliable or non-reproducible.

Instructions

Define the grader — express the acceptance criteria as atomic, deterministic rules (each independently checkable by code or a fixed script). No LLM-as-judge.
Fault injection (three classes):
- explicit — overt errors (timeouts, 500s, missing file, type mismatch).
- implicit — silent data degradation (truncated fields, missing fields, wrong-but-valid values).
- mixed — a combination of explicit and implicit faults.
Run each fault class against the output. Record, per rule, whether it held.
Emit a robustness profile — a table: Rule, Fault class, Held?, Evidence. Implicit faults carry higher weight (they lack overt error signals).

Output Contract

Per-rule pass/fail table with evidence for each verdict.
A single grader_score (fraction of rules held), sourced from the deterministic grader.
A list of the specific violated rules (trace each failure to a rule).

Constraints

Do NOT substitute LLM-as-judge for a deterministic grader.
Do NOT report a rule as "held" without running the check.
Implicit faults MUST be tested; happy-path-only is not a self-test.

---name: self-testdescription: Self-evaluates the agent's output under controlled fault injection using deterministic rule-based graders. Invoke to verify a deliverable before delivery, to test robustness against environmental degradation, or when acceptance must not rely on LLM-as-judge. Trigger phrases "test this", "verify my output", "check robustness", "self-test", or as the Verify stage of the metacognitive loop.---# self-testOccuBench-style fault injection + GDPevo-style deterministic grading. The goal is robustness over the "happy path", with a reproducible pass/fail per violated rule.## When to Use- Before delivering any artifact, to gate on deterministic criteria.- When the environment may degrade silently (APIs time out, data arrives incomplete).- When an LLM-as-judge would be unreliable or non-reproducible.## Instructions1. **Define the grader** — express the acceptance criteria as atomic, deterministic rules (each independently checkable by code or a fixed script). No LLM-as-judge.2. **Fault injection (three classes)**: - `explicit` — overt errors (timeouts, 500s, missing file, type mismatch). - `implicit` — silent data degradation (truncated fields, missing fields, wrong-but-valid values). - `mixed` — a combination of explicit and implicit faults.3. **Run each fault class** against the output. Record, per rule, whether it held.4. **Emit a robustness profile** — a table: `Rule`, `Fault class`, `Held?`, `Evidence`. Implicit faults carry higher weight (they lack overt error signals).## Output Contract- Per-rule pass/fail table with evidence for each verdict.- A single `grader_score` (fraction of rules held), sourced from the deterministic grader.- A list of the specific violated rules (trace each failure to a rule).## Constraints- Do NOT substitute LLM-as-judge for a deterministic grader.- Do NOT report a rule as "held" without running the check.- Implicit faults MUST be tested; happy-path-only is not a self-test.

Frequently asked questions