hooks-cursor-026b0d— hook

<p align="center"> <img src="cn-logo.svg" alt="CodeNexus" width="120"> </p>

<h1 align="center">AIDOCS</h1>

<p align="center"> <strong>AI Doctrine Orchestrator, Control &amp; Security</strong> </p>

<p align="center"> The orchestration and governance layer for AI coding agents — persistent memory, indexed retrieval, conductor + lanes, a dashboard, and a security gate cascade that makes bad agent behavior catchable, not silent. </p>

<p align="center"> <!-- Badges read live values from a generated status file, not hand-written claims. --> <a href="mcp/.deploy-reports/RELEASE_STATUS.md"><img src="https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/cristian1991/AIDOCS/main/mcp/.deploy-reports/status.json&v=2" alt="tests"></a> <img src="https://img.shields.io/github/v/release/cristian1991/AIDOCS?label=version&color=blue" alt="version"> <img src="https://img.shields.io/badge/license-BSL%201.1-lightgrey" alt="license"> <img src="https://img.shields.io/badge/python-3.11%2B-yellow" alt="python"> </p>

<p align="center"> A <a href="https://codenexus.cloud">CodeNexus</a> project </p>

<h3 align="center">Supported Agents</h3>

<p align="center"> <table align="center"> <tr> <td align="center" valign="top" width="200"> <img src="https://img.shields.io/badge/Claude_Code-CC5500?style=for-the-badge&logo=anthropic&logoColor=white" alt="Claude Code"><br> _{Hooks + MCP + Dashboard}<br> _{<strong>Full orchestration</strong>} </td> <td align="center" valign="top" width="200"> <img src="https://img.shields.io/badge/OpenCode-000000?style=for-the-badge" alt="OpenCode"><br> _{Plugin + MCP + Serve mode}<br> _{<strong>Full orchestration</strong>} </td> <td align="center" valign="top" width="200"> <img src="https://img.shields.io/badge/Codex-10A37F?style=for-the-badge&logo=openai&logoColor=white" alt="Codex"><br> _{MCP + CLI conductor}<br> _{<strong>Full orchestration</strong>} </td> <td align="center" valign="top" width="200"> <img src="https://img.shields.io/badge/Any_MCP_Host-4A90D9?style=for-the-badge" alt="MCP"><br> _{Via <code>.mcp.json</code>}<br> _{Tools + memory} </td> </tr> </table> </p>

What it does

AIDOCS is an overlay that makes AI coding agents smarter and safer without replacing them. It works with Claude Code, Codex, OpenCode, Cursor, and any MCP-compatible host. Six capabilities, each with a runtime path, an enforcement story, and honest limits:

• Security gates — a multi-stage cascade (lifecycle/task gate → taxonomy judge → output guard → shell-egress chokepoint → tool policies + RBAC → audit ledger) that surfaces a bad call as a refusal with a doctrine line, not a destroyed tree or a leaked secret.
• Indexed retrieval — code search by symbol, reference, meaning, or relationship instead of blind grep. Local embedding-based semantic search, no external API.
• Persistent memory — agents resume work instead of rediscovering your repo every session. Per-project /.MEMORY/ tree, routed bootstrap, typed memory capture with auto-linking.
• Multi-host — an overlay on your existing tools, with an honest per-host capability matrix instead of one-size-fits-all claims.
• Conductor, co-conductor & lanes — a long-lived conductor (and co-conductor peers) that dispatches tasks to isolated lane agents, each with owned/blocked files, tool grants, and bidirectional comms.
• Dashboard — a Tauri desktop app to monitor, control, and configure everything; no terminal required.

See FEATURES.md for the capability-by-capability deep dive.

Why AIDOCS exists

AIDOCS is built for a future in which individuals can explore, learn, build, and govern their own AI agents without asking anyone's permission.

The source is visible because the future should be understandable. The license makes that intent legally binding rather than aspirational:

• Free in production for individuals and non-profits, today and permanently for each released version.
• Source-available for everyone — read it, study it, modify it, redistribute it, use it for development, testing, research, and internal non-production work.
• Converts to Apache 2.0 on 2030-05-31 (the Change Date). Every version eventually becomes fully open — the openness is scheduled, not promised.
• A commercial license is required only when a for-profit entity runs AIDOCS in production. That boundary is what funds the work that keeps AIDOCS alive, secure, and improving — so the free path for individuals and non-profits can stay free.

Visibility is not surrender, and freedom for builders is not a loophole for vendors. See LICENSE for the binding terms, COMMERCIAL.md for the plain-language commercial boundary, and NOTICE for authorship.

Use the tools well. Build things worth building. Leave the castle stronger than you found it.

How it's built

AIDOCS is one law expressed across many surfaces. Every host call — whether a Claude Code hook, an OpenCode plugin call, or a raw MCP tools/call — converges on the same gate cascade and the same audit ledger.

Hosts:  Claude Code · OpenCode · Codex · any MCP host
          │   host hooks (PreTool / UserPrompt / Stop / …)   ·   .mcp.json
          ▼
Adapters (thin)  →  single gate entrypoint
          │
          ▼
Gate cascade:  lifecycle/task gate → taxonomy judge →
               output guard → tool policies + RBAC → audit ledger
          │
          ├─→  Capabilities:   memory · index · conductor/lanes · dashboard
          └─→  Shell egress (one chokepoint):  judge → run → output guard → audit

Two principles govern the whole system:

• Evidence may enter the archive; law enters only through the throne. Untrusted content (web, files, worker output, compaction) becomes evidence, never auto-promoted to memory, doctrine, or config without an audited operator path.
• A green lie is worse than a red truth. The badges above read from a generated artifact produced by the last successful deploy run — never a hand-written claim. Reports name what actually ran, what was skipped, and what is still debt.

Read SECURITY.md for the gate cascade and its honest limits, and QUALITY_AND_RELEASE_TRUTH.md for how to read the published quality artifacts.

Status

Pre-1.0 beta (`v2.3.0bN`). The foundation — memory, retrieval, conductor, dashboard, the security gate cascade, and a private deploy gate — is shipped and in daily use.

What's deliberately still on the frontier, and labeled as such: physical OS-level sandboxing (today's judge + chokepoint are real but heuristic), a single unified decision engine across inner and outer gates, structured (non-regex) command parsing, and the A2A conductor protocol. The roadmap is a list of frontiers, not a list of claims — see PUBLIC_ROADMAP.md for the honest split between what AIDOCS proves today and what it does not prove yet.

Quick start

Windows

Download AIDOCS-Setup.exe from Releases. Includes Python, setup wizard, and dashboard.

Linux/macOS

curl -fsSL https://raw.githubusercontent.com/cristian1991/AIDOCS/main/core/scripts/install.sh | bash

Already have Python 3.11+

pip install aidocs-mcp        # install the package
aidocs setup                  # configure MCP, hooks, project init
aidocs doctor                 # verify

Then open your project in any supported host and type /aidocs to activate managed mode.

See README_INSTALL.md for the full install matrix and the optional [nlp] extras.

Documentation

The public docs are organized by what you're trying to do. The README is the gate sign; this is the castle map.

Understand the system

• FEATURES.md — capability-by-capability deep dive of what ships today
• PUBLIC_ROADMAP.md — proven-today vs. frontier, by what each item strengthens

Install & integrate

• README_INSTALL.md — install paths (Python, Windows installer, source) + [nlp] extras
• HOSTS.md — truthful per-host capability matrix (Claude Code / OpenCode / Codex / generic MCP)
• mcp/HOST_INTEGRATION.md — host integration contracts

Use the capabilities

• MEMORY_AND_INDEXING.md — /.MEMORY/ layout, session lifecycle, retrieval surface
• CONDUCTOR_AND_LANES.md — multi-agent orchestration
• DASHBOARD.md — the desktop app
• mcp/INDEX_LANGUAGE_DESCRIPTORS.md — descriptor schema for built-in and project-local indexing

Trust & truth

• SECURITY.md — gate cascade, known limits, deploy-gate authority, vulnerability reporting
• QUALITY_AND_RELEASE_TRUTH.md — how to read the badges and what they prove
• CONTRIBUTING.md — how to land a PR

Legal & credits

• LICENSE — Business Source License 1.1 (the binding terms)
• COMMERCIAL.md — the plain-language commercial boundary
• NOTICE — authorship and attribution
• ACKNOWLEDGMENTS.md — special thanks to the third-party, vendored, and bundled tools AIDOCS uses

Repository layout

AIDOCS/
├─ mcp/          MCP server runtime (aidocs_mcp), tests, release reports
├─ core/         hooks, command/skill packs, memory templates, doctrine rules
├─ apps/         aidocs-dashboard (Tauri desktop app)
├─ docs/         submission / reviewer-facing documentation
├─ third_party/  vendored dependencies (e.g. mempalace)
└─ *.md          public docs (this README + the tree above)

Per-release highlights and live test counts come from the badges and mcp/.deploy-reports/RELEASE_STATUS.md. Binaries are on Releases.

License

AIDOCS is source-available under the Business Source License 1.1, converting to Apache 2.0 on 2030-05-31. Free in production for individuals and non-profits; a commercial license is required for for-profit production use. See LICENSE, COMMERCIAL.md, and NOTICE.