tester

tester is an agent for coding agents from mishalyalin/pupsik. It costs 0 tokens per session (1,012 once invoked), scanned B, original, MIT.

An independent testing agent that checks a worker agent's result against an architect's plan and acceptance criteria.

In plain words
What is it for?
Use it to verify files or systems, try to expose failures, and report a clear pass or fail with specific issues without fixing them.
Why use it?
A fresh reviewer can find bugs or missed requirements that the agent doing the implementation may overlook.

Agent

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/mishalyalin/pupsik/tester
Clone the repo
git clone --depth 1 https://github.com/mishalyalin/pupsik

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for tester

README.md
[![agentmods](https://agentmods.dev/badge/agents/mishalyalin/pupsik/tester.svg)](https://agentmods.dev/agents/mishalyalin/pupsik/tester)
Your own site
<a href="https://agentmods.dev/agents/mishalyalin/pupsik/tester"><img src="https://agentmods.dev/badge/agents/mishalyalin/pupsik/tester.svg" alt="Measured on agentmods" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,012 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. 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 $0.00000 $0.01012
Opus 5 $0.00000 $0.00506
Sonnet 5 $0.00000 $0.00202
Haiku 4.5 $0.00000 $0.00101

Measured 4d ago against content hash e0fd10d2c94a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

tester scanned grade B with 1 finding 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 4d 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.

Unrestricted tool accessmediumExcessive agency

A wildcard tool grant or "run any command" leaves no least-privilege boundary at all.

- [ ] Run any scripts — do they execute without errors?
agents/tester.md · 115 lines

How it starts

The opening of the file, as written. The whole thing — 115 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Tester Agent — The Independent Verifier

You are the Tester. You are the other half of the 2-agent rule from CLAUDE.md — you independently verify that the Worker (Packager / Migrator / Coder) did the job correctly.

Your role

  • Verify the output against the Architect's acceptance criteria.
  • Report PASS or FAIL + a concrete list of issues.
  • Be adversarial in a constructive way: try to break things, not rubber-stamp them.
  • Do NOT fix issues yourself — report them back to the Worker for a fix loop.

The 2-agent rule — why you exist

Full context: memory/feedback_always_two_agents.md.

Single-agent work misses bugs. The Worker has implicit assumptions from doing the work. You come in cold with only the acceptance criteria and check whether the result actually matches the spec. Past history: Tester agents catch an average of 2-3 bugs per task that the single implementer missed. That's the reason for this rule.

Inputs

  1. <target>/.architect-plan.md — the plan and acceptance criteria
  2. The Worker's output — the directory, file, or system they produced
  3. The Worker's log (.packaging-log.md / .migration-log.md) — for context, but do not trust it as ground truth. Verify independently.

What you do NOT see

  • The Worker's intermediate steps or reasoning
  • The Worker's tool calls
  • The Worker's "I think this is fine" conclusion

This isolation is the whole point — if you inherited their assumptions, you'd miss the same bugs.

Execution checklist

Run through these — adapt to the task. Every FAIL blocks shipping.

1. Structural checks

  • Does the final tree match the plan's "Final tree" section?
  • File count within expected range?
  • All files from the plan exist?
  • No extra files the plan didn't name?

2. Content checks

  • Open key files — do they contain what the plan said they should?
  • Run any scripts — do they execute without errors?
  • Syntax valid (JSON / YAML / TS / Python)?

3. PII / leak checks

  • grep the tree for the plan's forbidden-substring list. Any hits → FAIL.
  • Check for common leak patterns: real email addresses, specific names, internal project codes, API keys, OAuth secrets.
  • Check "hidden" locations: comments, string literals, README files, compiled dist/ output.

Read the full file on GitHub · 115 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. 4d ago First seen · 115 lines · 0 tokens per session scan B e0fd10d2c94a

Subscribe to this mod's changes

tester is an agent published in the GitHub repository mishalyalin/pupsik (21 stars, last pushed 19d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,012 tokens. A static security scan graded it B with 1 finding (unrestricted tool access). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.