Prisma AIRS Codex Hooks

Codex CLI hooks that scan prompts, MCP tool calls, and assistant responses through Prisma AIRS — blocking prompt injection, malicious code, DLP violations, and toxic content inside the agentic loop.

Docs: https://cdot65.github.io/prisma-airs-codex-hooks/ — architecture, configuration reference, hook contracts, and development guides all live there.

Coverage

| Prompt | Response | Pre-tool (MCP) | Post-tool (MCP) | Bash / apply_patch | | :-------: | :----------------------: | :------------: | :-------------: | :----------------: | | ✅ blocks | ⚠️ post-stream terminate | ✅ blocks | 👁 observe | — by design |

Setup

git clone https://github.com/cdot65/prisma-airs-codex-hooks.git
cd prisma-airs-codex-hooks
pnpm install && pnpm build

export PRISMA_AIRS_API_KEY="your-x-pan-token"
export PRISMA_AIRS_PROFILE_NAME="your-security-profile"

pnpm install-hooks              # project-level (<git-root>/.codex/)
pnpm install-hooks --global     # or user-level (~/.codex/)

Then open Codex, run /hooks, and trust the new hook definitions (required — trust is hash-based; re-trust after any reinstall).

Verify:

pnpm validate-connection
pnpm verify-hooks

Everything else

Installation options · Configuration (mode, fail_mode, profiles, content limits) · Hook contracts · CLI commands · Architecture · Contributing

MIT — see LICENSE.