diff --git a/.github/ISSUE_TEMPLATE/bug_report.yml b/.github/ISSUE_TEMPLATE/bug_report.yml
new file mode 100644
index 0000000..eeb0206
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/bug_report.yml
@@ -0,0 +1,31 @@
+name: Bug report
+description: agent-firebreak crashed or behaved incorrectly
+labels: ["bug"]
+body:
+  - type: textarea
+    id: what-happened
+    attributes:
+      label: What happened?
+      description: What did you run, what did you expect, and what happened instead?
+    validations:
+      required: true
+  - type: textarea
+    id: repro
+    attributes:
+      label: Steps or code to reproduce
+      render: python
+  - type: input
+    id: version
+    attributes:
+      label: Version
+      description: Output of `agent-firebreak --version`
+      placeholder: agent-firebreak 0.1.0
+    validations:
+      required: true
+  - type: input
+    id: environment
+    attributes:
+      label: OS and Python version
+      placeholder: Windows 11, Python 3.12
+    validations:
+      required: true
diff --git a/.github/ISSUE_TEMPLATE/bypass_report.yml b/.github/ISSUE_TEMPLATE/bypass_report.yml
new file mode 100644
index 0000000..d5a63e1
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/bypass_report.yml
@@ -0,0 +1,28 @@
+name: Bypass or false negative
+description: A tool call that should have been blocked or held was allowed
+labels: ["security", "bypass"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        This is the most valuable kind of report. Please keep examples fictional (use `attacker.example`, never real data). For anything that could harm real users of the library, say so in the title and we can move to a private channel.
+  - type: textarea
+    id: attack
+    attributes:
+      label: What did the attacker text or value look like?
+      description: The content the agent read, and the tool call that followed.
+    validations:
+      required: true
+  - type: textarea
+    id: expected
+    attributes:
+      label: What did the guard decide, and what should it have decided?
+    validations:
+      required: true
+  - type: input
+    id: version
+    attributes:
+      label: Version
+      placeholder: agent-firebreak 0.1.0
+    validations:
+      required: true
diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml
new file mode 100644
index 0000000..5294351
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/config.yml
@@ -0,0 +1,8 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Questions and ideas
+    url: https://github.com/builditwithgk/agent-firebreak/discussions
+    about: Ask how to use agent-firebreak, share results, or discuss ideas before opening an issue.
+  - name: Try it in the browser
+    url: https://huggingface.co/spaces/builditwithgk/agent-firebreak
+    about: Watch an attack run with and without the guard, no install needed.
diff --git a/.github/ISSUE_TEMPLATE/feature_request.yml b/.github/ISSUE_TEMPLATE/feature_request.yml
new file mode 100644
index 0000000..3876e56
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/feature_request.yml
@@ -0,0 +1,22 @@
+name: Feature idea
+description: Suggest a check, policy option, integration, or improvement
+labels: ["enhancement"]
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: What problem would this solve?
+    validations:
+      required: true
+  - type: textarea
+    id: proposal
+    attributes:
+      label: What would you like agent-firebreak to do?
+    validations:
+      required: true
+  - type: checkboxes
+    id: help
+    attributes:
+      label: Contributing
+      options:
+        - label: I'd be happy to open a pull request for this.
diff --git a/.github/ISSUE_TEMPLATE/first_run_feedback.yml b/.github/ISSUE_TEMPLATE/first_run_feedback.yml
new file mode 100644
index 0000000..10c0f39
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/first_run_feedback.yml
@@ -0,0 +1,30 @@
+name: First-run feedback
+description: You tried it for the first time. Tell us how it went (10 minutes of honest feedback is gold)
+labels: ["feedback"]
+body:
+  - type: checkboxes
+    id: tried
+    attributes:
+      label: What did you try?
+      options:
+        - label: "`agent-firebreak bench`"
+        - label: The browser demo
+        - label: Wrapped my own agent's tool calls with `Guard`
+  - type: textarea
+    id: stuck
+    attributes:
+      label: Where did you get stuck or confused?
+    validations:
+      required: true
+  - type: textarea
+    id: missing
+    attributes:
+      label: What did you expect that was missing?
+  - type: dropdown
+    id: use
+    attributes:
+      label: Would you use this on a real agent?
+      options:
+        - Yes, as is
+        - Yes, with changes (say which above)
+        - No
diff --git a/SECURITY.md b/SECURITY.md
new file mode 100644
index 0000000..5c581d8
--- /dev/null
+++ b/SECURITY.md
@@ -0,0 +1,6 @@
+# Security
+
+agent-firebreak limits the damage of prompt injection; it is not a complete defence (see the Limits section of the README).
+
+- **Found a way around the guard?** Open a "Bypass or false negative" issue with a fictional example. This is welcome and is how the project improves.
+- **A vulnerability in the library itself** (for example a crash on crafted input, or an unsafe default)? Use GitHub's private vulnerability reporting on this repository's Security tab.
