Mistral Vibe: The /loop Command for Recurring Agent Prompts

Mistral Vibe

Mistral Vibe v2.9.5 introduced the /loop command, enabling developers to run any prompt or slash command on a recurring interval directly within the CLI — turning one-off agent tasks into repeating routines without external scripting. The release also added the default_agent configuration option, allowing users to persistently set their preferred agent profile in config.toml instead of passing --agent flags on every launch. Several stability improvements accompany the release, including fixes for parallel-instance history conflicts, session ID display, and restored default TUI scroll sensitivity.

Sources & Mentions

2 external resources covering this update


The /loop Command: Recurring Agent Automation

Mistral Vibe v2.9.5 ships the /loop command, a new built-in that allows users to run any prompt or slash command on a repeating interval directly within the CLI. Where previously automating repetitive agent tasks required external scripting or cron jobs, /loop makes recurring execution a first-class primitive inside the Mistral Vibe session.

The practical range of use cases is broad: a developer can use /loop to periodically re-run a test suite check, re-invoke a custom skill to watch for regressions, poll for a specific condition in a codebase, or chain together repeating code review passes during an active development session. Any prompt or slash command that works interactively now also works as a looped routine.

Default Agent Configuration

Also introduced in this release is the default_agent configuration option. Prior to v2.9.5, users who preferred a specific agent profile — such as plan for a read-only exploration mode or accept-edits for auto-approving file changes — had to specify it explicitly on each launch. The default_agent setting in config.toml persists that preference globally (or per-project when placed in .vibe/config.toml), eliminating the need to pass --agent flags on every invocation.

Valid values include any of the built-in profiles (default, plan, accept-edits, auto-approve, lean when installed) or the name of any custom agent file in ~/.vibe/agents/ or the project's .vibe/agents/ directory. The setting applies only to interactive sessions; in programmatic mode (-p / --prompt), Mistral Vibe continues to fall back to auto-approve when no --agent flag is passed.

Stability and Refinements

v2.9.5 also includes a set of reliability improvements:

  • Session IDs in compacted sessions now display correctly
  • The history file is reloaded before writing, preventing parallel Vibe instances from overwriting each other's history
  • read_file is correctly flagged as truncated when the file size limit is reached
  • Default scroll sensitivity in the TUI is restored to its expected value
  • The enable_telemetry config key now correctly takes precedence over the older enable_otel key
  • The "no git commit" constraint is loosened to accommodate more workflows
  • CLI loader alignment is corrected

Additionally, telemetry instrumentation was extended to cover the teleport command, and environment variable logging options are now documented in --help output, making Mistral Vibe's diagnostic surface more discoverable.


Mentioned onGitHubDeepWiki