feat: add HOL Guard workflow safety skill

.claude-plugin/marketplace.json

@@ -11,13 +11,14 @@
   "plugins": [
     {
       "name": "reflex",
-      "description": "Skills for the Reflex web framework — docs, environment setup, and process management",
+      "description": "Skills for the Reflex web framework — docs, environment setup, process management, and guarded agent workflows",
       "source": "./",
       "strict": false,
       "skills": [
         "./skills/reflex-docs",
         "./skills/setup-python-env",
-        "./skills/reflex-process-management"
+        "./skills/reflex-process-management",
+        "./skills/reflex-hol-guard"
       ]
     }
   ]

skills/reflex-hol-guard/SKILL.md

@@ -0,0 +1,55 @@
+---
+name: reflex-hol-guard
+description: >
+  Protect agent-driven Reflex development and operations with HOL Guard on the local
+  coding-agent boundary. Use when an agent will run shell, package, Git, database,
+  cloud, or deployment commands for a Reflex app and runtime policy, approvals, or
+  evidence are needed before state-changing tool execution.
+---
+
+# Reflex Workflow Safety with HOL Guard
+
+Use HOL Guard as a safety layer around the **local coding-agent harness** that is working on a Reflex project. HOL Guard does not run inside the Reflex web server and this skill does not claim server-side Reflex interception.
+
+Keep using the repository's other Reflex skills for framework-specific work: `reflex-docs` for current APIs, `setup-python-env` before project setup, and `reflex-process-management` for compile/run/reload/debug workflows.
+
+## Start the agent through HOL Guard before mutations
+
+Install the maintained CLI if needed:
+
+```bash
+pipx install hol-guard
+```
+
+Identify the local agent harness instead of guessing a name:
+
+```bash
+hol-guard detect
+```
+
+Use the exact identifier returned by `detect` to install protection and launch a protected session:
+
+```bash
+hol-guard install <detected-harness>
+hol-guard run <detected-harness>
+```
+
+If protection cannot be established, do not describe the current session as guarded. Keep remaining work read-only or restart from a HOL Guard-protected session before state-changing calls.
+
+Before package installs, Git writes, database migrations, cloud changes, deploys, process termination, or other mutating commands, verify the setup:
+
+```bash
+hol-guard doctor
+```
+
+Command inspection is inspection only. A command-test result is not proof that the current agent process is intercepted; the protected harness launched through HOL Guard is the enforcement boundary.
+
+## Preserve Reflex-native safeguards
+
+HOL Guard is additive. Do not bypass normal Reflex and platform controls because Guard is present. Keep least-privilege credentials, normal Git review, database backups, cloud IAM, preview/dry-run operations, and project validation. If Guard denies or holds an action for review, do not execute the blocked operation through an unguarded shell or alternate tool.
+
+## Completion checks
+
+Before claiming a protected mutation completed, confirm that the coding-agent session was launched through HOL Guard before the mutation, the state-changing operation went through that protected harness, deny/review/error outcomes prevented execution, and Reflex-native validation still ran where applicable.
+
+For runtime details and supported harnesses, use the maintained HOL Guard documentation at https://github.com/hashgraph-online/hol-guard.