You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This report covers the last 24 hours of firewall activity across all agentic workflows in github/gh-aw that use the firewall feature. Data was aggregated from 56 workflows across 87 firewall-enabled runs (50 additional run summaries were skipped as malformed/missing and are excluded from this analysis).
Overall firewall activity was light: of 4,184 total monitored network requests, only 208 (≈5%) were blocked. The vast majority of blocked traffic is concentrated in a small number of domains — notably a Sentry ingest endpoint (likely a false-positive candidate for allowlisting) and several Google/consumer-web domains triggered almost exclusively by the "Smoke Codex" workflow. No signs of malicious or suspicious blocked traffic were observed; blocked domains largely reflect legitimate services (telemetry, GitHub, npm, Go proxy) that are either intentionally restricted or missing from allowlists.
📊 Key Metrics
Metric
Value
Total network requests monitored
4,184
✅ Allowed requests
3,976
🚫 Blocked requests
208
Block rate
4.97%
Total unique blocked domains
17
Terminology Note: Allowed requests successfully reached their destination; blocked requests were prevented by the firewall. A low block rate with several listed blocked domains indicates most traffic is legitimate, with occasional blocked calls to services outside the configured allowlist.
🚫 Top Blocked Domains
Domain
Times Blocked
Workflows
Category
o205451.ingest.us.sentry.io:443
87
Agent Container Smoke Test, Code Scanning Fixer, Contribution Check, Daily Model Inventory Checker, Deployment Incident Monitor, Design Decision Gate 🏗️, Documentation Unbloat, PR Sous Chef, Smoke Checkout PR Dispatch, Smoke Kiro, Smoke Multi PR, Smoke Pi, Smoke Project, Smoke Service Ports, Smoke Temporary ID, Test Quality Sentinel
Analytics/Tracking
github.com:443
27
Contribution Check, Design Decision Gate 🏗️, PR Sous Chef, Test Quality Sentinel
Development Services
ab.chatgpt.com:443
19
Auto-Triage Issues, Avenger, Bot Detection, Daily Go Test Parallelizer, Daily Observability Report for AWF Firewall and MCP Gateway, Ponytail Reviewer, Smoke Call Workflow, Smoke Codex, Smoke OTEL
Note: Chart generation was attempted via the firewall-chart-generator sub-agent but it returned no output/URLs, so trend charts are omitted from this report.
Allowlist Sentry ingest endpoint: o205451.ingest.us.sentry.io:443 is by far the most frequently blocked domain (87 hits) yet is also allowed thousands of times in the same runs (e.g., 94 allowed vs 22 blocked in PR Sous Chef). This pattern (mixed allow/block for the same domain) suggests intermittent or per-request policy enforcement rather than a hard block — worth reviewing the relevant workflows' network permissions to ensure consistent allowlisting of this legitimate telemetry endpoint.
github.com:443 blocked in several workflows (Contribution Check, Design Decision Gate, PR Sous Chef, Test Quality Sentinel) while api.github.com:443 and githubcopilot.com are allowed — these workflows likely need github.com (non-API) added to their network allowlist if they perform browser-style GitHub requests (e.g., fetching raw pages or redirects).
api.anthropic.com:443 is blocked a handful of times but overwhelmingly allowed (e.g., 102 allowed vs 1 blocked in Smoke Claude) — likely transient/retry blocks rather than a policy gap; no action needed beyond monitoring.
ab.chatgpt.com:443 blocked consistently across many workflows (19 total, 9 workflows) with zero allowed requests — this looks like an OpenAI CDN/analytics domain not required for API functionality; blocking appears intentional and correct.
Smoke Codex is a clear outlier: it alone accounts for the majority of Google/consumer-web blocked domains (www.google.com, accounts.google.com, content-autofill.googleapis.com, android.clients.google.com, clients2.google.com, www.gstatic.com, host.docker.internal:4173). These appear to be background browser/Chrome telemetry calls from an embedded browser tool used in that smoke test — likely benign, but confirm the smoke test doesn't need real Google service access; if not, no allowlist change is required.
Policy rule data unavailable: no policy_analysis field was present in any of the analyzed run summaries, so rule-level attribution (Section 4) could not be generated for this report. Consider adding policy rule metadata to firewall log capture if per-rule effectiveness tracking is desired.
No suspicious/malicious domains observed — all blocked domains map to known legitimate services (GitHub, npm, Go proxy, Sentry, Anthropic, OpenAI/ChatGPT, Google/Chrome telemetry, Cursor).
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
🔥 Executive Summary
This report covers the last 24 hours of firewall activity across all agentic workflows in
github/gh-awthat use the firewall feature. Data was aggregated from 56 workflows across 87 firewall-enabled runs (50 additional run summaries were skipped as malformed/missing and are excluded from this analysis).Overall firewall activity was light: of 4,184 total monitored network requests, only 208 (≈5%) were blocked. The vast majority of blocked traffic is concentrated in a small number of domains — notably a Sentry ingest endpoint (likely a false-positive candidate for allowlisting) and several Google/consumer-web domains triggered almost exclusively by the "Smoke Codex" workflow. No signs of malicious or suspicious blocked traffic were observed; blocked domains largely reflect legitimate services (telemetry, GitHub, npm, Go proxy) that are either intentionally restricted or missing from allowlists.
📊 Key Metrics
🚫 Top Blocked Domains
Note: Chart generation was attempted via the
firewall-chart-generatorsub-agent but it returned no output/URLs, so trend charts are omitted from this report.View Detailed Request Patterns by Workflow
Workflow: PR Sous Chef (1 run analyzed)
Workflow: Smoke Codex (1 run analyzed)
Workflow: Smoke Claude (1 run analyzed)
Workflow: Design Decision Gate 🏗️ (1 run analyzed)
Workflow: Test Quality Sentinel (1 run analyzed)
Workflow: Contribution Check (1 run analyzed)
Workflow: Code Scanning Fixer (1 run analyzed)
Workflow: Smoke Cursor (1 run analyzed)
Other workflows with blocked traffic (1 domain each, 1 run analyzed)
View Complete Blocked Domains List
🛡️ Security Recommendations
o205451.ingest.us.sentry.io:443is by far the most frequently blocked domain (87 hits) yet is also allowed thousands of times in the same runs (e.g., 94 allowed vs 22 blocked in PR Sous Chef). This pattern (mixed allow/block for the same domain) suggests intermittent or per-request policy enforcement rather than a hard block — worth reviewing the relevant workflows' network permissions to ensure consistent allowlisting of this legitimate telemetry endpoint.github.com:443blocked in several workflows (Contribution Check, Design Decision Gate, PR Sous Chef, Test Quality Sentinel) whileapi.github.com:443andgithubcopilot.comare allowed — these workflows likely needgithub.com(non-API) added to their network allowlist if they perform browser-style GitHub requests (e.g., fetching raw pages or redirects).api.anthropic.com:443is blocked a handful of times but overwhelmingly allowed (e.g., 102 allowed vs 1 blocked in Smoke Claude) — likely transient/retry blocks rather than a policy gap; no action needed beyond monitoring.ab.chatgpt.com:443blocked consistently across many workflows (19 total, 9 workflows) with zero allowed requests — this looks like an OpenAI CDN/analytics domain not required for API functionality; blocking appears intentional and correct.www.google.com,accounts.google.com,content-autofill.googleapis.com,android.clients.google.com,clients2.google.com,www.gstatic.com,host.docker.internal:4173). These appear to be background browser/Chrome telemetry calls from an embedded browser tool used in that smoke test — likely benign, but confirm the smoke test doesn't need real Google service access; if not, no allowlist change is required.policy_analysisfield was present in any of the analyzed run summaries, so rule-level attribution (Section 4) could not be generated for this report. Consider adding policy rule metadata to firewall log capture if per-rule effectiveness tracking is desired.All reactions