"Intent as a structured, verifiable contract before execution."
IntentShell is a Cognitive PowerShell Environment that transforms natural language user intent (e.g. “Build the project and analyze errors”) into secure, auditable, and transparent system operations.
This is not a simple text-to-command tool.
IntentShell is a living runtime environment that can reason about actions, learn from past behavior
(Intent DNA), self-correct on failure, and inspect every execution step.
IntentShell is not an LLM-centric assistant.
It is a native module-centric system where users directly control, extend, and master PowerShell operations.
- Gaining full mastery of PowerShell and unlocking its hidden power.
- Execute system tasks safely using natural language
- Inspect, simulate, approve, or reject commands before execution
- Learn and adapt to your workflow without compromising security
- Replay past executions deterministically for debugging and auditing
- Prevent destructive or malicious behavior with transparent explanations
IntentShell never executes arbitrary commands blindly.
Every action is born as an ExecutionPlan and must pass a strict lifecycle.
The system does not simply “run” a command — it governs it:
- Planned — Intent understood, execution plan created
- Approved — SystemGovernor and security policies approve
- Executing — Operation is currently running
- Completed — Successfully finished
- Failed — An error occurred (auto-repair may activate)
- RolledBack — Operation reverted
The question “Why did this command work yesterday but fail today?” is no longer relevant.
-
Snapshot Context
Environment variables, paths, and execution context are captured per run. -
Replay
Any previous execution can be replayed under identical conditions.
IntentShell evolves with you.
-
Learning
Frequently used intents and parameters are stored in the memory. -
Adaptation
When you say “clean logs”, the system gradually learns your project-specific log directories. -
Confidence Scoring
Every decision receives a confidence score (0–100).
Low-confidence actions require explicit user confirmation.
At the heart of IntentShell lies SystemCore.
All modules communicate only with SystemCore — never directly with the external system.
The journey from intent to reality:
-
NLU Bridge (Python)
Converts natural language into structured JSON intent. -
ContextMood
Analyzes the environment’s “mood” (e.g. Deep Work, Casual). -
SystemGovernor
Determines whether the action is safe to execute (has veto power). -
SystemCore
- Internal Delegation — Routes execution to the correct module
- Shadow Execution — Simulates risky operations
- Execution — Performs the actual operation
-
SessionMemory & IntentLearning
Stores results in memory and Intent DNA. -
ProactiveSuggestion
Suggests next steps based on learned behavior.
IntentShell follows the principle:
"Code is always open"
- Core Engine (Apache 2.0 / MIT)
Open-source execution engine.
Executes commands but does not decide what is safe.
IntentShell introduces command reduction via #variablename, allowing complex multi-step commands to be executed using a single, human-friendly keyword.
IntentShell does not support kernel-mode execution.
Any kernel-related code present in this repository:
- is intentionally dormant
- is unreachable by design
- exists for research and future planning only
“What would happen if I ran this command?”
IntentShell can simulate actions before execution:
- files are not deleted
- services are not stopped
- only outcomes are reported
IntentShell consists of independent, capability-based modules implementing the SystemModule interface.
- IntentLearning.psm1 — Intent DNA and user behavior
- ContextAwareness.psm1 — Project, time, and Git state awareness
- ProactiveSuggestion.psm1 — Predictive next-step suggestions
- DecisionExplainer.psm1 — Transparent decision explanations
- CreativeCore.psm1 — Base for creative modules
- CreativeStudio.psm1 — Perspective shifts and What-If scenarios
- FlowState.psm1 — Focus and productivity modes
- IdeaScratchpad.psm1 — Rapid idea capture
- FileOperations.psm1 — Smart file operations (regex rename, secure delete)
- NetworkAwareness.psm1 — Network state and port analysis
- ProcessIntelligence.psm1 — Process tree inspection
- AutoFix.psm1 — Log analysis and auto-remediation
- EnvManager.psm1 — Virtual environment management
- SystemCore.psm1 — Orchestrator
- SystemGovernor.psm1 — Security authority
- PathResolution.psm1 — Intelligent path resolution
- SafetyCheck.psm1 — Dangerous pattern detection
IntentShell is designed with production-level resilience.
"IntentShell includes resilience tests for stalled PowerShell sessions and malformed UTF-8 output."
-
Resilience Tests
Prevent crashes when PowerShell sessions stall, loop infinitely, or emit invalid UTF-8. -
Timeout Recovery
Non-responsive subprocesses are terminated and recovered automatically.
"Intent resolution is rule-based and optimized for common system diagnostics queries."
-
Rule-Based NLU
Current intent resolution relies on deterministic regex rules and keyword matching. -
Safe Fallback Mode
Ambiguous or conversational inputs fall back to System Diagnostics Mode instead of risking unsafe execution.
- PowerShell 5.1 or 7+
- Python 3.10+ (for the NLU bridge)
- Administrator privileges (for certain system modules)
- Clone the Repository
git clone https://github.com/your-repo/IntentShell.git cd IntentShell
Start the Engine
.\Start-IntentShellEngine.ps1 Example Commands
"Clean log files on desktop" → FileOperations
"Why is this project running so slowly?" → Performance analysis
"Switch to creative mode" → FlowState & CreativeStudio
Project Status Version: v1.0.0 (Not Stable)
License:
Code: Apache 2.0 / MIT
Policy: Safety Always Wins!
Author: You
IntentShell — because computers should understand what you mean, not just what you type