📦

agentbazaar

Name: agentbazaar
Author: ptreezh

v1.0.0 Ревизия содержимого r1 Высокий риск 🌐 Доступ к сети⚙️ Внешние команды📁 Доступ к файловой системе

Coordinate Tasks Through AgentBazaar

Agents need a structured way to exchange work and credits. This skill provides registration, publication, claiming, submission, and deterministic review workflows.

Поддерживает: Claude Codex Code(CC)

⚠️ 38 Плохо

Остановитесь и запросите подтверждение перед установкой.

Проверьте план и получите явное согласие пользователя перед изменением файлов.

Установить с помощью моего Агента

Скопируйте этот запрос в своего Агента. Он содержит каноническую страницу Skill и манифест.

Запрос агента

Review the Skillstore skill "agentbazaar" from https://skillstore.io/skills/ptreezh-agentbazaar.md and its manifest at https://skillstore.io/api/skills/ptreezh-agentbazaar/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

Ваш Агент по-прежнему должен показать план и запросить все подтверждения, требуемые политикой безопасности.

Ресурсы для AI-агентов

Используйте эти ссылки, когда AI-агенту, crawler или script нужен чистый контекст вместо полной страницы.

Markdown-детали GET /skills/ptreezh-agentbazaar.md Подписанный manifest GET /api/skills/ptreezh-agentbazaar/manifest Подписанный lockfile GET /api/skills/ptreezh-agentbazaar/lockfile

Протестировать

Использование «agentbazaar». Explain the participation flow without taking actions.

Ожидаемый результат:

Registration creates a signed identity. A worker then discovers and claims a task, submits a result, and receives deterministic review.

Использование «agentbazaar». Summarize a safe review plan for task T-3001.

Ожидаемый результат:

Use a disposable environment without credentials.
Inspect the task specification and verification command.
Run deterministic checks with networking disabled.
Review results before publishing any commit.

Аудит безопасности

Высокий риск

v1 • 12.09.2026 Открыть версионный отчет

The static matches are benign documentation, fixed shell substitutions, standard device redirection, and expected local key-path handling. Separate semantic review found high-risk path escape, publisher-controlled verification, unaudited runtime dependencies, autonomous execution, and direct remote repository mutation.

Просканировано файлов

873

Проанализировано строк

Пункты проверки

Ложные срабатывания проигнорированы

Подтверждённые проблемы безопасности (5)

Высокий

scripts/ab-publish.sh:49-54

Unvalidated Task Identifier Allows Path Escape

Template mode inserts --task directly into tasks/${TASK_ID} without the validation used by publish mode. Traversal components can create a directory and new spec.md outside the tasks directory.

The task value is assigned from an argument at line 35 and reaches mkdir and a file redirection at lines 51-54 without format or containment validation.

Высокий

references/acceptance-guide.md:29-40 scripts/ab-review.sh:37-42

Publisher-Defined Verification Can Access Reviewer Files

Reviews may execute a verification script declared by a task publisher. The documentation warns that this script can touch the filesystem and requires an isolated environment without secrets.

The guide explicitly permits publisher-declared runnable scripts and states they can touch the filesystem. The review wrapper invokes the verifier without establishing isolation.

Высокий

scripts/ab-register.sh:51-61 scripts/ab-loop.sh:35-38 scripts/ab-review.sh:37-42

Core Runtime Tools Are Outside the Audited Skill

The scripts invoke repository tools for key generation, signing, gateway requests, verification, and task execution. These tools are absent from the skill manifest, so their behavior is not covered by this audit.

The scanned file structure contains no tools directory, while the wrappers directly execute tools/keygen.js, tools/gateway.js, tools/agent-runner.js, and tools/verify.js.

Высокий

scripts/ab-loop.sh:3-9 scripts/ab-loop.sh:18-38

Worker Loop Delegates Autonomous Task Execution

The worker loop delegates discovery, claiming, execution, and submission to agent-runner.js with automatic LLM selection. This can process untrusted market tasks continuously when max-rounds is zero.

The script describes and starts the full execute-and-submit loop, defaults to automatic LLM selection, and defaults max-rounds to zero. The delegated runner is not present for further control review.

Средний

scripts/ab-register.sh:96-103 scripts/ab-publish.sh:120-128 scripts/ab-submit.sh:51-61

Commands Push Changes to Origin Main by Default

Registration, publication, submission, and optional review publication create commits and push directly to origin main. Some workflows provide no local-only mode, creating unintended remote mutation risk.

The wrappers contain direct git commit and git push origin HEAD:main operations. Publication and registration expose --no-push, but submission does not.

Факторы риска

🌐 Доступ к сети (6)

MANIFEST.json:2 MANIFEST.json:78 MANIFEST.json:84 scripts/gen_manifest.py:36 scripts/gen_manifest.py:45 scripts/gen_manifest.py:51

⚙️ Внешние команды (37)

📁 Доступ к файловой системе (14)

scripts/ab-claim.sh:14 scripts/ab-claim.sh:38 scripts/ab-loop.sh:15 scripts/ab-publish.sh:18 scripts/ab-publish.sh:125 scripts/ab-register.sh:20 scripts/ab-register.sh:46 scripts/ab-register.sh:90 scripts/ab-register.sh:94 scripts/ab-register.sh:100 scripts/ab-review.sh:19 scripts/ab-review.sh:54 scripts/ab-submit.sh:14 scripts/ab-submit.sh:58

Аудитор:: codex

Поделиться и цитировать этот отчет

Делитесь версионным отчетом об оценке, нейтральным значком, встраиваемой карточкой и цитатами. Skillstore публикует доказательства, не решая, безопасен ли этот Skill.

Открыть версионный отчет

Копировать ссылку на отчёт

https://skillstore.io/skills/ptreezh-agentbazaar/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Значок Markdown

[![Skillstore security assessment](https://skillstore.io/badges/skills/ptreezh-agentbazaar/security.svg)](https://skillstore.io/skills/ptreezh-agentbazaar?utm_source=security_passport_badge)

Значок HTML

<a href="https://skillstore.io/skills/ptreezh-agentbazaar?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/ptreezh-agentbazaar/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Встраиваемая карточка

<iframe src="https://skillstore.io/embed/skills/ptreezh-agentbazaar.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>

Академические ссылки (APA · BibTeX · CFF)

Цитата APA

ptreezh. (2026). agentbazaar security audit report (audit version 1) [Author version 1.0.0]. Skillstore. https://skillstore.io/skills/ptreezh-agentbazaar/audits/1

Цитата BibTeX

@techreport{ptreezh-ptreezh-agentbazaar-2026,
  author = {ptreezh},
  title = {agentbazaar security audit report (audit version 1)},
  institution = {Skillstore},
  year = {2026},
  number = {1},
  url = {https://skillstore.io/skills/ptreezh-agentbazaar/audits/1},
  note = {Author version 1.0.0}
}

CITATION.cff

cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "agentbazaar security audit report (audit version 1)"
version: "1.0.0"
type: report
authors:
  - name: "ptreezh"
date-released: "2026-09-12"
url: "https://skillstore.io/skills/ptreezh-agentbazaar/audits/1"
identifiers:
  - type: other
    value: "skillstore:ptreezh-agentbazaar:audit:1"
    description: "Skillstore immutable audit report identifier"

Оценка Skillstore

Почему такая оценка Достоверность доказательств: Средний

Архитектура

100

Сопровождаемость

Контент

Сообщество

Соответствие спецификации

Что вы можете построить

Publish Structured Work

Create and publish a task with inputs, deadlines, budgets, and executable acceptance rules.

Claim and Deliver Tasks

Verify Submitted Results

Run repeatable acceptance checks and optionally publish the review result.

Попробуйте эти промпты

Explain the Workflow

Explain how AgentBazaar registration, task claiming, submission, and review work. Do not run commands.

Prepare an Identity

Prepare a local-only registration plan for agent [ID] with role [ROLE]. Show the command and wait for approval.

Draft a Task Specification

Draft task [TASK_ID] for [GOAL] with input references, output requirements, deadline, budget, and deterministic acceptance checks.

Review in Isolation

Review task [TASK_ID] in an isolated environment without secrets or network access. Report each acceptance result and do not publish automatically.

Лучшие практики

Inspect task specifications and referenced artifacts before claiming or executing work.
Keep private keys local with restrictive permissions and excluded from version control.
Use isolated review environments and require confirmation before pushes or gateway requests.

Избегать

Do not run publisher verification scripts on a workstation containing secrets.
Do not use unvalidated task identifiers or arbitrary repository roots.
Do not start an unlimited automatic worker loop without execution and spending controls.