LIVE
Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|Google Introduces "Gemini 4 Carbon" as Next Step in Its Model Lineup28/10/26 · Google|DeepMind unveils Genie 4, a new generation of world model27/10/26 · Google DeepMind|Anthropic files IPO prospectus, with risk disclosures taking center stage26/10/26 · Anthropic|Robot policy trained only in simulation matches top real-world systems26/10/26|OpenAI unveils GPT-6.1 Terra, a model built for computer use26/10/26 · OpenAI|Google's AI gives evasive answers about who won the 2020 US election28/10/26 · Google|NVIDIA introduces Nemotron 3.5 Transcribe, a promptable multilingual transcription model28/10/26 · NVIDIA|OpenAI and ServiceNow deepen partnership on enterprise workflows28/10/26 · OpenAI|Failed startups' data finds a second life training AI models28/10/26|Anthropic expands AI for Science program to 20 foundations and research institutions28/10/26 · Anthropic|Anthropic shares early results on using Claude to accelerate drug development27/10/26 · Anthropic|GEPA-Curator: Filtering Failures to Better Train LLM Agents27/10/26|
Toolofficial publisher

Permissions (Claude Code)

Tool : A precise action the AI can call · requires claude-code

Allow, deny, and ask rules per tool and per pattern.

The simplest safeguard for a code agent, to be configured before anything else.

Docs
Overall score
8.2 /10
Security: nothing to flag
Utility
8.0
Reliability
6.0
Security
No caveat on record
Ease
5 min · Comfortable

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

When a code agent can read, modify, and execute files without restriction, the risk of an unwanted action is real. Claude Code's permission system addresses this by letting you define precisely what the agent is allowed to do, tool by tool and command by command.

In practice, each rule follows the format Tool or Tool(specifier) and can be of type allow, deny, or ask. Deny rules are evaluated first, then ask, then allow: the first match wins. A deny rule on a bare tool name removes that tool from the agent's context entirely, while a scoped rule like Bash(rm *) keeps the tool available but blocks matching calls.

These rules are stored in versionable settings.json files, allowing a team to share a common configuration while letting each developer adjust their own preferences. The /permissions dialog lets you view and modify rules mid-session, even while an agent is working.

This tool is aimed at developers using Claude Code who want to constrain the agent's actions without disabling its capabilities. It is not relevant for those who do not use Claude Code or who prefer to let the agent work without restrictions.

What it offers

  • ·Three rule types (allow, deny, ask) configurable per tool and per command or path pattern.
  • ·Rules evaluated in a fixed order: deny first, then ask, then allow, regardless of rule specificity.
  • ·Complete removal of a tool from the agent's context via a bare-name deny rule, without a specifier.
  • ·Per-repository persistence of approvals for Bash commands and WebFetch domains, saved to .claude/settings.json.
  • ·/permissions dialog accessible mid-session to add or remove rules with immediate effect on the next tool call.
  • ·Ability to attach a comment when answering a permission prompt, forwarded to Claude as context.
  • ·Several permission modes available: manual, acceptEdits, plan, auto, dontAsk, bypassPermissions, with the option to disable auto and bypassPermissions via configuration files.
  • ·Input parameter matching for deny and ask rules, for example Agent(model:opus) or Bash(run_in_background:true).

In practice

Price
Free
Commercial use
Commercial use: yes
Available on
claude-code
Host
claude-code
FR interface
No

Security and data

Rating
Security: nothing to flag
Capabilities
readswritesexecutesnetwork
Your data
Data stays local
Maintenance
Active

Links and repository checked on 25 October 2026 · added on 1 February 2025

How to install Permissions (Claude Code)

1. Install Claude Code and open a git repository in your terminal. 2. Launch Claude Code in that repository so rules are associated with the correct context. 3. Type /permissions in the interface to open the rule management dialog. 4. Add your allow, deny, or ask rules using the format Tool or Tool(specifier). 5. Verify that the .claude/settings.json file at the repository root contains your rules before sharing them with your team. 6. For shared environments, set permissions.disableBypassPermissionsMode to "disable" in the configuration file to prevent any bypass.
.claude/settings.json → permissions

Official documentation ↗

In the same category

Whole category Monitor →