面向 Agent SaaS 访问的统一权限与审计层

一个统一控制平面,将 Agent 的身份映射为最小权限范围,并对 Jira、GitHub、Slack 等 SaaS 的访问进行完整审计,填补操作系统层到应用层之间的权限缺口。

I want to build a venture-scale product. Help me plan and start the MVP.

Product: Unified permission and audit layer for agent SaaS access
What it is: One control plane mapping an agent's identity to least-privilege scopes and full audit across Jira, GitHub, Slack and other SaaS, closing the OS-to-app permission gap.
Who pays: Security/platform leads at 200–5,000 person companies deploying coding and ops agents; $2–10k/mo from the identity/security budget.
Closest alternatives: Microsoft Entra Agent ID / Agent 365 (Microsoft-stack only, early), Okta / Auth0 (human-centric, no agent-aware resource models), Spera/Oasis (non-human identity posture, no per-action policy enforcement), Manual per-tool API token scoping (status quo)
The gap: OS sandboxes contain the process but are blind to what the agent does inside SaaS APIs; existing IdPs model humans, not per-task agent scopes. No neutral, cross-SaaS layer enforces 'read prod config, never write it' at the API level with unified audit.
Why now: Enterprises are deploying agents with broad OAuth tokens right now (2025-26), regulators and SOC teams are asking for agent attribution, and Agent 365's arrival validates the category while leaving non-Microsoft shops open.
Moat to build toward: Per-SaaS resource-model mappings and normalized audit schemas are slow to replicate; once it is the system of record for agent permissions, rip-and-replace cost is high.

可直接粘贴到 Claude Code、Cursor 或任何编程助手的需求说明（英文）。