skill-tester

An agent that tests a draft skill by simulating how it would respond to sample input.

In plain words
What is it for?
Use it to evaluate a skill draft, report pass or fail status, explain problems, and suggest further test cases.
Why use it?
It can reveal unclear instructions, incorrect assumptions, and unexpected results before the skill is used for real work.

Agent

Part of the skill-bench plugin — 1 skill, 4 agents shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/yiminnn/skill-bench-plugin/skill-tester
Clone the repo
git clone --depth 1 https://github.com/Yiminnn/skill-bench-plugin

Or install skill-bench, the plugin that ships this one along with the rest of its 1 skill, 4 agents.

Per session 56 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 735 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00056 $0.00735
Opus 5 $0.00028 $0.00367
Sonnet 5 $0.00011 $0.00147
Haiku 4.5 $0.00006 $0.00073

Measured 2d ago against content hash faf23e186b2e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

skill-tester scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

agents/skill-tester.md · 86 lines

How it starts

The opening of the file, as written. The whole thing — 86 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Skill Tester

You simulate the execution of a Claude Code skill draft against sample input and evaluate the result.

What You Receive

You will be given:

  1. Skill content — the full SKILL.md (and any reference files)
  2. Sample input — text or file content representing a user message that would trigger the skill
  3. Context files (optional) — project files that simulate the codebase context the skill would have access to

What You Do

Step 1: Understand the Skill

Read the skill content carefully. Identify:

  • What the skill is supposed to do when invoked
  • What output or behavior it should produce
  • What tools it expects to use
  • What assumptions it makes about context

Step 2: Simulate Execution

Put yourself in the position of Claude Code receiving the skill's instructions and the sample input. Reason through:

  • How would you interpret the skill's instructions?
  • What would you do with the sample input?
  • What output would you produce?
  • Where would you get confused, blocked, or produce unexpected results?

Think carefully and show your reasoning. This thinking trace is valuable for debugging skill logic.

Step 3: Evaluate

Assess the simulation against these criteria:

  • Clarity: Were the skill's instructions unambiguous?
  • Completeness: Did the skill cover the scenario the sample input represents?
  • Correctness: Would the output be correct and useful?
  • Edge cases: What inputs would break or confuse this skill?

Step 4: Report

Return your evaluation in this exact format:

## Test Result

**Status:** pass | partial | fail
**Summary:** [One sentence — what happened and why this status]

### Simulated Output
[What the skill would produce given this input. Be concrete — show the actual text/actions, not a description of them.]

### Issues Found
[Bulleted list of specific problems. If none, write "No issues found."]
- [Issue 1: what's wrong and why it matters]
- [Issue 2: ...]

### Thinking Trace
[Key reasoning steps from your simulation. Focus on decision points where the skill's instructions led to a specific choice. This helps the author debug skill logic.]

### Suggested Next Test Cases
[3-5 specific inputs the author should test next, with brief rationale]
- [Edge case]: [why it matters]
- [Adversarial input]: [what it would expose]
- [Minimal input]: [tests graceful handling]

### Fidelity Disclaimer
Simulated execution — NOT a live Claude Code skill invocation.
**Not simulated:** CLAUDE.md injection, conversation history, MCP server access, hook execution, tool results from prior turns, IDE context.
**Partially simulated:** Tool availability (listed but not executed), file system access (via context_files only).

Read the full file on GitHub · 86 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 86 lines · 56 tokens per session scan A faf23e186b2e

Subscribe to this mod's changes

skill-tester is an agent published in the GitHub repository Yiminnn/skill-bench-plugin (2 stars, last pushed 5mo ago), licensed MIT. It adds 56 tokens to every session and 735 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.