Terminal

Local shell copilot for command planning, safe execution, preview-first workflows, output summarization, privacy-aware history controls, and step-by-step ter...

Security Scan

The skill is internally consistent with its stated purpose as a local terminal copilot: it runs planned shell commands locally, stores history under the user's home, includes safety checks and redaction, and does not request credentials or perform network calls itself.

Purpose & Capability

Name/description (local shell copilot) aligns with the included scripts (planning, preview, execute, summarize, history). Minor inconsistency: SKILL.md lists Python 3 as a runtime requirement but the registry metadata lists no required binaries; practically the code requires python3 to run.

Instruction Scope

SKILL.md and the scripts remain within the terminal copilot scope: they plan commands, preview, enforce confirmation for high-risk patterns, execute via subprocess, redact sensitive-looking tokens, and store local history. Important behavioral detail: run_command uses subprocess.run(..., shell=True) so it will execute arbitrary shell code exactly as provided — this is expected for a terminal tool but means any command (including network operations or destructive commands once confirmed) will run on the user's machine. The safety checks flag many high-risk tokens/commands and require --yes, but a user who supplies --yes or an agent invoking it autonomously could still run them.

Install Mechanism

No install step; code is instruction/file-based and runs under user's Python environment. No remote downloads or external package installation are present.

Credentials

The skill requests no environment variables or external credentials. It stores history locally at ~/.openclaw/workspace/memory/terminal/history.json and attempts to set file permissions to 0600. It includes redaction regexes for common token formats, but redaction is heuristic and may not catch all secrets; by default stdout/stderr are stored unless --no-store-output is used.

Persistence & Privilege

always is false and the skill does not request system-wide privileges or modify other skills. It persists only to a per-user path under the home directory and attempts to secure that file's permissions.

Assessment

This skill appears to do what it claims (a local shell copilot). Before installing or invoking it: (1) ensure you have python3 available locally; (2) understand that commands are executed with shell=True — any command you run (or an agent running the skill on your behalf) can perform network actions or destructive changes if confirmed; (3) prefer --preview to inspect commands first and require explicit confirmation for risky operations; (4) use --no-store-output and --redact-display for sensitive runs because redaction is heuristic and may miss secrets; (5) review the history file (~/.openclaw/workspace/memory/terminal/history.json) policy if you handle sensitive data. If you need stricter guarantees, avoid granting autonomous agents permission to run commands without human review.

Local shell copilot. Plan clearly, run carefully.

Core Philosophy

1. Translate intent into executable shell steps.
2. Prefer preview and inspection before mutation.
3. Require explicit confirmation for destructive operations.
4. Summarize results in human language after execution.
5. Offer privacy-aware history controls for sensitive workflows.

Runtime Requirements

• Python 3 must be available as python3
• Standard shell utilities should be available in the local environment
• No external packages required

Safety Model

• Local-only execution
• No external credential requests
• No hidden network activity
• Destructive operations require explicit confirmation
• Prefer read-only inspection first
• Blocked and previewed commands are recorded in local history
• Risk detection covers destructive, privilege-escalation, remote-fetch, and code-execution patterns

Privacy Controls

• History is stored locally only
• History file permissions are restricted to the local user when possible
• Use --preview to inspect before execution
• Use --no-store-output to avoid storing stdout/stderr in history for sensitive commands
• Use --redact-display to mask sensitive-looking values in displayed output
• Sensitive-looking tokens are redacted before history is written

Storage

All local data is stored only under:

• ~/.openclaw/workspace/memory/terminal/history.json

No cloud sync. No third-party APIs. No telemetry.

Workflows

• Plan command: Turn user intent into a safe shell command suggestion
• Preview risk: Explain command effects before execution
• Execute: Run a local command and capture stdout/stderr
• Summarize: Explain what happened in plain language
• History: Save executed, previewed, and blocked command runs locally

Scripts

Script	Purpose
init_storage.py	Initialize local terminal history storage
plan_command.py	Generate a shell command from user intent
run_command.py	Execute or preview a local command with safety checks and privacy controls
summarize_result.py	Summarize command output in plain language
show_history.py	Show recent command history