ToShop DocsTerminal
Run shell commands inside a task, scoped and audited.
The Terminal skill (part of the Developer Tools bundle) lets your agent run shell commands as part of a task — with the same approval and audit gates as every other tool.
Enable it
Install Developer Tools
Settings → Skills → Developer Tools → Install.
Enable for your agent
Toggle the Terminal capability on for the agent that needs it (typically your Dev Agent).
Approve first commands as they come
The first time your agent runs each new command pattern, you'll be prompted. Approvals can be one-time or persistent.
How it works
Managed shell session
State preserved across calls within a task — cd, env vars, etc. carry over.
Approval per command pattern
"Run git status" might be auto-approved after the first time; "run rm" always asks regardless.
Streaming output
Output streams back into chat as it produces — you see it in real time.
Shell environment
- Shell — Picks up your default shell (zsh, bash, or fish) from your login env.
- PATH — Inherited from your login shell environment.
- Long-running processes — Managed via the underlying tmux session — they keep running between your agent's steps.
sudo— Always requires explicit approval per invocation, with a separate confirmation.
When to use it
One-off shell tasks
"Count last 30 days of orders in orders-export.csv by region" — things you'd otherwise type yourself.
Local scripts
"Run the sync-shopify-prices.sh script and tail the log until it finishes."
Batch operations
Where the shell is the easiest tool — find files by date, grep order IDs across CSVs, jq JSON inventory exports.
For repeatable shell workflows, wrap them in a Skill so your agent can pick the right command for the goal.
Audit log
Every shell call lands in Settings → Local Tools with the command, exit code, and duration. Sensitive arguments (recognized secrets) are redacted in the log.
Related
- Skills — Developer Tools bundle, custom skills.
- Permissions — managing approval rules for shell patterns.