Why YeePilot

Built for Real Workflows

YeePilot is a Go-based CLI/TUI agent for coding and shell workflows. It is designed for execution-heavy tasks where safety and traceability matter.

Stage-Driven Agent Runtime

The runtime follows explicit stages:

1. discover
2. plan
3. edit/execute
4. verify
5. review
6. finalize

A clarify stage is used when requirements are ambiguous.

This structure keeps complex tasks predictable and auditable.

Controlled Autonomy

Autonomy is configurable per session/profile:

• high
• guarded
• strict_review
• ultra

Use /autonomy status to inspect active behavior and recovery guards.

Verification + Recovery by Default

YeePilot is not only generation and execution. It includes a bounded recovery system with:

• post-mutation verification evidence
• failure classification (patch_conflict, build_breakage, test_regression, verification_failed)
• budgeted retry/repair cycles
• patch-first auto-repair path and conflict-aware rebase apply

This is built for completing tasks reliably, not just producing output.

Freshness and Knowledge Layer

YeePilot includes a built-in freshness layer (with configurable policy) and optional Context7 integration:

• ai.knowledge.*
• ai.context7.*

This helps keep technical answers aligned with current official docs and releases.

Security-Centered Execution

Core safeguards include:

• command validation and risk classification
• security modes (strict, moderate, permissive)
• sandbox controls
• tamper-aware audit logging with integrity checks

Local Encrypted Vault

YeePilot Vault stores secrets locally in encrypted form and supports multiple unlock methods:

• password
• pin
• yubikey
• yubikey+pin
• recovery key

Vault includes autolock and brute-force protections, plus SSH trust workflows.

Provider Flexibility

Use OpenRouter, OpenAI, or Anthropic and switch as needed without changing your operational workflow.

Bottom Line

YeePilot combines modern agent execution with controls expected in serious environments: staged planning, explicit approval boundaries, verification/recovery loops, and local secret protection.