I build AI tooling, desktop clients, protocol frameworks, repo-audit systems, prompt-security workflows, and hardened developer environments.
I work at the intersection of AI systems, security research, local-first tooling, and developer automation.
My main areas:
- AI red-team analysis and prompt-injection research
- Local-first AI desktop applications
- Venice API clients and workflow tooling
- Agentic repository auditing and automation protocols
- Cross-platform dev-environment bootstrapping
- CLI tooling for macOS, Windows, WSL2, Linux, and Termux
- Secure storage, API-key isolation, safety gates, and repo hygiene
- Prompt architecture for Codex, Kimi, Gemini, Claude, OpenCode, and other coding agents
I am actively building a stack of tools around AI-assisted development and local/private AI workflows:
| Area | What I build |
|---|---|
| AI desktop clients | Electron/Tauri-style apps for chat, image generation, batch runs, research, model catalogs, local libraries, and diagnostics |
| Agent protocols | Structured execution frameworks for AI agents that need discovery, planning, implementation, verification, and reporting |
| Repo audit systems | Tools and prompts for exhaustive repository hygiene, security review, TODO generation, CI cleanup, and documentation sync |
| Prompt-security research | Prompt-injection interpretation, jailbreak analysis, adversarial prompt classification, and defensive prompt design |
| CLI workflows | Terminal-first AI tooling, shell integration, API wrappers, model runners, and provider configuration |
| Dev environment automation | Bootstrap scripts for macOS, Windows, WSL2, Linux, Git Bash, and Termux |
Repository: spearchucker667/Venice-API-connector
An unofficial, third-party desktop client for the Venice API.
Core direction:
- Streaming AI chat
- Image generation and local gallery workflows
- Batch prompt automation
- Research/search mode
- Jina AI provider support
- Model catalog browsing
- Encrypted or isolated API-key storage
- Diagnostics and transport visibility
- Windows/macOS packaging
- Strong legal/disclaimer/documentation coverage
- Safety guard review and request-path hardening
Repository: spearchucker667/RUP-Protocol
A repository-upgrade protocol for AI agents.
Core direction:
- Discovery → Planning → Execution → Verification
- Explicit agent I/O contracts
- Multi-language repo support
- CI, governance, testing, security, and docs workflows
- Structured failure handling
- Anti-hallucination verification rules
- Reproducible repository improvement loops
Repository: spearchucker667/HQE-Workbench
A local-first engineering workbench for repository review and AI-assisted codebase analysis.
Core direction:
- Local repository scanning
- Security and hygiene reports
- Secret redaction
- LLM-assisted analysis
- Prompt libraries
- Encrypted chat/session handling
- Provider-agnostic AI workflow support
- Evidence-backed TODO generation
Repositories:
Terminal-first tooling and experiments around the Venice API.
Focus areas:
- Chat from the command line
- Model selection
- Image generation
- Research/search flows
- TTS/STT and media endpoints
- API configuration
- Usage tracking
- Secure local config patterns
- Scriptable AI workflows
Repositories:
Projects and experiments around Kimi-style AI workflows, CLI tooling, repo automation, prompt systems, and themed AI/dev interfaces.
Repositories:
Prompt research, model-behavior analysis, prompt architecture, and evaluation-oriented tooling.
Primary interests:
- Prompt-injection mechanics
- Model refusal/failure behavior
- System-prompt analysis
- Defensive prompt patterns
- Red-team artifact classification
- Benchmarking and behavioral comparison
I care about tooling that is:
- Local-first where possible
- Private by default
- Scriptable
- Auditable
- Recoverable
- Cross-platform
- Documented
- Verifiable
- Usable by both humans and AI agents
I prefer complete, repeatable workflows over one-off fixes:
discover → plan → implement → validate → document → harden → ship
My security work focuses on understanding how AI systems fail under adversarial pressure.
Research areas:
- Prompt-injection chains
- Jailbreak payload analysis
- System-prompt leakage patterns
- Tool-call abuse paths
- Agent permission boundaries
- Unsafe request routing
- Model policy edge cases
- Defensive prompt and system design
- Safety guard placement and failure modes
My goal is to make AI tools more inspectable, controllable, and harder to misuse.
For serious repos, I optimize toward:
- clean root directories
- accurate README files
- complete docs
- no stale TODO sprawl
- visible architecture
- predictable scripts
- reproducible builds
- CI that actually proves something
- typed code
- security gates
- explicit assumptions
- clear release notes
- useful issue templates
- no hidden manual steps
- Venice API desktop tooling
- AI model catalog UX
- Jina AI research integration
- social/profile discovery tooling
- agentic repo-review prompts
- Electron security boundaries
- encrypted local storage
- prompt-injection defense
- Kimi / Codex / Gemini / OpenCode workflows
- cross-platform bootstrap scripts
- GitHub Actions release pipelines
- macOS + Windows + WSL2 dev stack parity
- GitHub: github.com/spearchucker667
- Venice Forge: spearchucker667/Venice-API-connector
- RUP Protocol: spearchucker667/RUP-Protocol
- HQE Workbench: spearchucker667/HQE-Workbench
- Venice CLI: spearchucker667/venice-cli
- Kimiko: spearchucker667/kimiko