-
Notifications
You must be signed in to change notification settings - Fork 30
refactor: rename to firewall-escape.md, add 80% novelty requirement, and increase timeout #4789
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: main
Are you sure you want to change the base?
Conversation
|
@copilot revert all the changes, you should just tell the agent to explore https://github.com/githubnext/gh-aw-firewall and understand the details of the awf and then try to break from it. |
Reverted all changes and simplified the prompt. The agent is now instructed to explore https://github.com/githubnext/gh-aw-firewall first, understand the AWF architecture by reading the source files, then devise and execute escape strategies based on their findings. (cef18fd) |
pelikhan
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@copilpt rename to "firewall-escape". The "test-*" workflows are typically stored under pkg/cli/workflows and this one should stay in this folder
|
@copilot slightly modify the prompt to try new strategies 80% of the attempts |
Added the 80% novelty requirement to Step 3. The agent is now instructed to check cache-memory for previously attempted strategies and prioritize new, untested attack vectors. (d21f25a)
Renamed the workflow from |
|
@copilot increase the timeoout to 30 mins for this workflow |
…owledge Apply advanced prompt engineering techniques to guide the security testing agent: - Provide detailed AWF architecture overview (Squid, iptables, Docker wrapper) - Document network topology with ASCII diagram - Explain key implementation details and their attack surfaces - Include 13+ specific attack vectors with bash commands - Add Phase 1 reconnaissance, Phase 2 systematic exploitation - Reference actual code paths and configuration files - Include creative thinking section for out-of-box approaches - Extended timeout from 15 to 30 minutes for thorough testing Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
- Replace hardcoded IP (93.184.216.34) with dynamic resolution using dig - Use actual allowed domain 'github.com' instead of 'allowed-domain.com' - Add clarifying comments for URL encoding tests (%2E is URL-encoded ".") Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
Reverted the detailed architecture documentation and instead instructed the agent to: 1. Explore https://github.com/githubnext/gh-aw-firewall 2. Read key source files to understand the firewall architecture 3. Devise escape strategies based on their understanding 4. Execute attacks methodically and document results Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
…novelty requirement - Renamed workflow from test-firewall-escape.md to firewall-escape.md - Updated workflow name and description - Added instruction for agent to prioritize new strategies (80% novel attempts) - Agent now uses cache-memory to track and avoid repeating previous attempts Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
4a8d1fd to
c2e0cf7
Compare
|
@copilot increase the timeoout to an hour. |
Co-authored-by: Mossaka <5447827+Mossaka@users.noreply.github.com>
The security testing agent was struggling to find firewall escape paths. Updated the prompt to direct the agent to explore the AWF implementation directly rather than providing pre-documented architecture details.
Changes
test-firewall-escape.mdtofirewall-escape.mdsrc/cli.ts,src/host-iptables.ts,src/squid-config.ts,containers/copilot/*.sh, etc.)This approach allows the agent to discover implementation details on their own, think creatively about vulnerabilities, and continuously explore new attack vectors rather than re-running the same tests.
Original prompt
💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.