The outcome

Create an auditable permission and sandbox baseline that behaves safely in interactive and non-interactive sessions.

CodeBuddy is a AI coding environment from Tencent Cloud with IDE, agent, and command-line workflows. Running complex coding tasks with visible artifacts and diffs while combining layered allow, ask, and deny rules with directory trust and OS-level shell sandboxing. This guide narrows that broad capability into one repeatable outcome, with checkpoints that keep the source material and your judgment in the loop.

Before you begin

Set the boundary before the tool starts.

Choose one real task, identify who will use the result, and decide what evidence or test will make the result acceptable. Gather only the source material needed for that task. If the work contains confidential, personal, regulated, or client-owned information, confirm that the platform and account are approved before sharing it.

Troiana principle

AI should make the work easier to inspect. If the workflow removes the source, the owner, or the review step, redesign the workflow.

Step by step

A workflow you can repeat.

  1. 01

    Inventory approved directories, commands, domains, MCP tools, models, subagents, credentials, environments, and actions that always require a person.

  2. 02

    Put shared least-privilege ask and deny rules in reviewed project settings while keeping machine-specific paths and credentials in uncommitted local or user scope.

  3. 03

    Disable bypass and auto modes where policy requires, enable sandbox filesystem and network allowlists, and disable unsandboxed escape when unnecessary.

  4. 04

    Test trusted and untrusted clones, protected files, dangerous command variants, redirects, chained commands, prompt injection, subagents, and non-interactive fallbacks.

  5. 05

    Audit effective rules in the permission panel, version owner-approved settings, rotate secrets, and rerun the test matrix after every CodeBuddy upgrade.

Working standard

What good use looks like.

  • Never commit credentials.
  • Test effective rules from an untrusted clone.
  • Deny unsandboxed escape unless required.

Directory trust, tool permissions, permission mode, and shell sandbox are separate controls. Use explicit deny rules as hard boundaries, keep bypass permissions disabled, prevent unsandboxed retries where not needed, restrict additional directories and network domains, review project settings before trusting a clone, cap parallel agents, keep credentials short-lived, and independently verify all code, tests, commands, commits, pushes, and deployments.

Official references

Check the current product documentation.

Features, plan limits, availability, and data controls change. These official pages are the starting points used for this collection.