Skip to content
Agentmart
Submit a tool

agentclash-challenge-pack-tools-sandbox

Use when defining AgentClash challenge pack tool access, sandbox runtime needs, filesystem expectations, network policy, command execution, and secret…

Skill by agentclash

IndexedMIT★ 36
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/agentclash-agentclash-challenge-pack-tools-sandbox

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use when defining AgentClash challenge pack tool access, sandbox runtime needs, filesystem expectations, network policy, command execution, and secret references.

Use cases: security

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/agentclash-challenge-pack-tools-sandbox

What's inside

Skill name
agentclash-challenge-pack-tools-sandbox
Description
Use when defining AgentClash challenge pack tool access, sandbox runtime needs, filesystem expectations, network policy, command execution, and secret references.
License
MIT
Metadata
agentclash.role: challenge-pack-tools · agentclash.version: 1 · agentclash.requires_cli: true
Instructions
240 lines in SKILL.md · view on GitHub

Files (1)

  • SKILL.md

SKILL.md (first lines)

# AgentClash Challenge Pack Tools And Sandbox

## Purpose
Define the native execution surface a challenge pack needs: pack-defined custom tools, broad tool policy, sandbox network/package/env settings, and safe secret references.

Use this skill only when a pack truly needs native files, tools, network, packages, or sandbox behavior. Keep the runtime surface narrow enough that failures are attributable to the agent, not to an over-broad environment.

## Use When
- A challenge pack needs top-level `tools.custom`.
- The pack needs `version.tool_policy.allowed_tool_kinds`.
- The pack needs `version.sandbox` for network access, CIDR allowlists, environment variables, apt packages, or a sandbox template.
- A coding agent needs exact source-backed YAML shapes without reading the AgentClash source repo.
- A reviewer needs to check that no raw secrets or unsupported tool kinds are being introduced.

## Do Not Use When
- The pack is `prompt_eval` and only needs prompt/final-output evaluation.
- The task is workspace infrastructure setup with `agentclash infra tool ...`; use `agentclash-runtime-resources-setup`.
- The task is artifact declaration, scoring validators, LLM judges, validation/publish, or eval running; use the focused downstream skills.

## Environment
Use hosted production for CLI examples unless the user intentionally targets a local or self-hosted backend.

```bash
export AGENTCLASH_API_URL="https://api.agentclash.dev"
```

## Validation Commands
Validate after adding or changing tools, tool policy, or sandbox settings.

```bash
agentclash challenge-pack validate path/to/pack.yaml
agentclash challenge-pack validate path/to/pack.yaml --json
```

Human output prints `Challenge pack is valid` or `Challenge pack has errors`. Use `--json` for structured `valid` and `errors` fields.

## Execution Mode Rules
`prompt_eval` packs cannot use challenge-pack tools or sandbox settings.

Do not include these in `prompt_eval`:
- top-level `tools`
- `version.tool_policy`
- `version.sandbox`

Use `native` when the task needs files, tool calls, network policy, extra packages, sandbox templates, file validators, directory checks, or code execution.

```yaml
version:
  number: 1
  execution_mode: native
```

## Tool Policy
`version.tool_policy.allowed_tool_kinds` accepts only these broad kinds:

```yaml
version:
  tool_policy:
    allowed_tool_kinds:
      - browser
      - build
      - data
      - file
      - network
```

Supported values are exactly `browser`, `build`, `data`, `file`, and `network`. Do not use `shell`; the current validator rejects it.

Use the narrowest set possible:
- `file` for reading/writing workspace files.
- `build` for build/test style operations.
- `network` for outbound HTTP or API access.
- `browser` for browser interaction.
- `data` for structured data access tools.

## Pack-Defined Custom Tools
Challenge-pack custom tools live at top-level `tools.custom`, not under `version`.

```yaml
tools:
  custom:
    - name: check_inventory
      description: Check inventory for a SKU.
      parameters:
        type: object
        properties:
          sku:
            type: string
        required:
          - sku
      implementation:
        primitive: http_request
        args:
          method: GET
          url: "https://api.example.com/inventory/${sku}"
          headers:
            Authorization: "Bearer ${secrets.INVENTORY_API_KEY}"
```

Source-backed fields:
- `tools.custom[]` entries are the supported pack-defined tool shape.
- `name` should be stable and unique in the pack.
- `parameters` must be valid JSON Schema when provided. If omitted, validation defaults to an empty object schema, but authoring explicit parameters is clearer.
- `implementation` is required.
- Non-`mock` implementations require `implementation.primitive`.
- Non-`mock` implementations require `implementation.args`, and `args` must be a JSON/YAML object.
- `implementation.primitive` cannot equal the tool's own `name`.
- Tool delegation cycles are rejected, and delegation depth greater than 8 is rejected.

Mock tools are the only exception to primitive/args validation:

```yaml
tools:
  custom:
    - name: fake_lookup
      parameters:
        type: object
      implementation:
        type: mock

From agentclash/agentclash, MIT.

More from agentclash/agentclash

Signals

Stars
36
Forks
3
Open issues
61
Last commit (this folder)
Jun 24, 2026 (4 months ago)
Last release
Aug 15, 2026 (2 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from agentclash

Cite this page

agentclash-challenge-pack-tools-sandbox skill. Agent Mart. https://agentmart.dev/skills/agentclash-agentclash-challenge-pack-tools-sandbox (last updated 2026-10-24)