Add policyengine-github-agent skill with doom loop prevention

.claude-plugin/marketplace.json

@@ -194,6 +194,7 @@
         "./skills/tools-and-apis/policyengine-python-client-skill",
         "./skills/tools-and-apis/policyengine-simulation-mechanics-skill",
         "./skills/tools-and-apis/policyengine-core-skill",
+        "./skills/tools-and-apis/policyengine-github-agent-skill",
         "./skills/domain-knowledge/policyengine-us-skill",
         "./skills/domain-knowledge/policyengine-uk-skill",
         "./skills/tools-and-apis/policyengine-api-skill",

skills/README.md

@@ -77,6 +77,7 @@ Knowledge about PolicyEngine's core tools, APIs, and applications.
 | **policyengine-api-skill** | PolicyEngine API | Flask REST service powering policyengine.org |
 | **policyengine-app-skill** | PolicyEngine React web application | User interface at policyengine.org |
 | **policyengine-core-skill** | PolicyEngine Core simulation engine | The foundation powering all PolicyEngine calculations |
+| **policyengine-github-agent-skill** | GitHub bot agent | Automated bot for issue/PR responses, avoiding doom loops |
 | **policyengine-python-client-skill** | Python client usage | Programmatic access via Python or REST API |
 
 ### 📊 Data Science (`data-science/`)

skills/tools-and-apis/policyengine-github-agent-skill/SKILL.md

@@ -0,0 +1,52 @@
+---
+name: policyengine-github-agent-skill
+description: Guidance for working with the PolicyEngine GitHub agent bot
+---
+
+# PolicyEngine GitHub Agent Skill
+
+## For Users 👥
+
+The PolicyEngine GitHub agent is an automated bot that can be invoked on GitHub issues and pull requests using `@policyengine` mentions. It helps with code reviews, bug fixes, and implementing features across PolicyEngine repositories.
+
+## For Contributors 💻
+
+### Critical: Avoiding Doom Loops
+
+**NEVER use '@policyengine' or '@policyengine-auto' in bot responses.** These mentions trigger the bot and create infinite loops where the bot repeatedly responds to itself.
+
+#### Why This Matters
+
+When the bot posts a comment containing `@policyengine`, GitHub notifies the bot account, which triggers another invocation. This creates a chain reaction of bot responses that can generate dozens or hundreds of comments before being stopped. See [PR #22](https://github.com/PolicyEngine/policyengine-github-agent/pull/22) for a real example of this issue.
+
+#### Safe Alternatives
+
+Instead of mentioning the bot directly:
+
+❌ **Don't do this:**
+- "Thanks @policyengine for the suggestion!"
+- "I've addressed @policyengine's feedback"
+- "cc @policyengine-auto"
+
+✅ **Do this instead:**
+- "I've implemented the suggested changes"
+- "The feedback has been addressed"
+- "Updated based on the review comments"
+- Simply refer to "the bot" or "the agent" without the @ mention
+
+### Bot Invocation
+
+The bot can only be invoked by members of the [PolicyEngine/core-developers](https://github.com/orgs/PolicyEngine/teams/core-developers) team. Non-members will receive a permission error.
+
+### Response Style
+
+When the bot responds to issues or PRs:
+- Keep responses concise and actionable
+- Focus on the technical task at hand
+- Avoid unnecessary pleasantries or acknowledgments
+- Never include @ mentions of the bot usernames
+
+## Resources
+
+- [policyengine-github-agent repository](https://github.com/PolicyEngine/policyengine-github-agent)
+- [PR #22: Stop doom loops](https://github.com/PolicyEngine/policyengine-github-agent/pull/22)