SoftSensorAI — a CLI-first way to make AI assistants actually productive for engineering teams
TL;DR. SoftSensorAI is a CLI-first toolkit that standardizes how teams use AI assistants (Claude, Codex, Gemini, Grok, Copilot) across skill levels and project phases. It ships with a two-tier install model, a policy engine for skills and phases, 60+ productionized command workflows, multi-persona capability packs, and security-aware CI gates. It turns “prompting” into repeatable engineering practice.
The link to the github repo — https://github.com/Softsensor-org/SoftSensorAI
Why this exists
AI coding tools are powerful but inconsistent: different teammates, machines, and repos produce different results. SoftSensorAI gives your org a stable interface to AI assistance, making outcomes repeatable across laptops, CI, and repos. Think “infrastructure for prompting.”
What it is (in one paragraph)
A global tool + per-repo configuration system. Globally, you install scripts, wizards, and agent templates once. Per repo, you add project-specific instructions, capability packs (personas), and command catalogs. The result is a predictable, policy-driven AI workflow that scales from solo devs to larger teams.
Core ideas, briefly
Two-tier install model. Global layer (~/softsensorai, toolchain, agent configs) + per-project layer (.claude/, AGENTS.md, scripts). This separation makes rollouts safe and reversible.
Skills & Phases = Policy engine. People vary (beginner→expert) and projects evolve (poc→mvp→beta→scale). SoftSensorAI changes capabilities and CI strictness accordingly (e.g., coverage and security gates tighten as you move to beta/scale).
Multi-Persona system. Composable expertise packs (data-scientist, software-architect, backend-developer, devops, security, etc.) that add domain-specific commands and guidance to your repo.
Command catalog (60+). Pre-built, auditable workflows: think-hard, explore-plan-code-test, backend-feature, security-review, tickets-from-code, long-context map-reduce, and more. These aren’t one-off prompts; they’re disciplined mini-playbooks with success criteria and validation.
Security & CI gates that escalate with risk. Default CI is light; project phases enable stricter checks: gitleaks, semgrep, trivy, coverage thresholds, and license checks — so noise doesn’t drown you at MVP, and real guardrails kick in before prod.
Get Vivek Gupta’s stories in your inbox
Join Medium for free to get updates from this writer.
GPU-aware AI stack (optional). When you need ML, SoftSensorAI detects NVIDIA/AMD/Apple GPUs, installs CUDA/ROCm-compatible stacks, and supports LangChain/AutoGen/CrewAI, vector DBs, and sandboxed execution with checksum-verified downloads.
Quick start (3 steps)
# 1) Global install (one time)
git clone https://github.com/Softsensor-org/SoftSensorAI.git ~/softsensorai
cd ~/softsensorai && ./setup_all.sh
export PATH="$HOME/softsensorai/bin:$PATH"# 2) Make a repo AI-ready
dp setup # paste your GitHub URL or run inside an existing repo# 3) Work with AI
ss review # AI code review
ss tickets # backlog from code
ss help
(Team/Server? Use the multi-user installer.)
How it works (under the hood)
- Prompt layering merges global→repo→task into
system/active.md, and all supported CLIs point to it. This creates consistent “brains” for AI tools—every time. - Persona stacks turn capabilities on/off per project type (e.g., architect+backend+devops for APIs; add security for microservices).
- Daily flow: profile/persona check → explore-plan-code-test → security-review → tickets-from-code → PR. The catalog is designed to be read-only safe by default, then escalate.
A week with SoftSensorAI (example cadence)
- Day 1 (Setup):
setup_all.sh, repo wizard, explore commands. - Day 2 (Feature):
/explore-plan-code-test+ security review. - Day 3 (Architecture): add architect persona, run
/architecture-review. - Day 4 (Quality): test-first pattern + audit.
- Day 5 (Tickets/Docs): generate tickets and sync API contracts.
Who should use this
- Teams wanting policy-driven AI workflows (skills/phases) with consistent results across repos and laptops.
- ML groups that benefit from GPU detection, optional AI frameworks, and sandboxed execution.
- Organizations that need explicit CI gates and license/compliance checks tied to maturity.
Roadmap & contributing
Docs and guide rails are comprehensive, but we’re expanding command docs, test coverage (BATS/integration matrix), and consistent error handling/rollback paths. Issues are labeled for newcomers and enhancements; PRs follow Conventional Commits.
License
SoftSensorAI is released under GNU GPL v3. If you share or modify, keep notices, provide the license text, and share source to recipients under the same terms. No warranty is provided.
Get started: ss setup, then run /explore-plan-code-test in your AI assistant. Star the repo and open an issue if you hit rough edges.
