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
Date: 2026-09-06 (report covers the last 24 hours of workflow activity)
This report analyzed 261 firewall-enabled workflow runs across 95 distinct workflows in the github/gh-aw repository. Overall firewall activity was dominated by an extreme outlier: a single workflow (PR Code Quality Reviewer) accounted for over 500,000 blocked requests to registry.npmjs.org:443 across just 7 runs — this appears to be a retry/polling loop rather than normal traffic and skews the aggregate block rate heavily. Outside of that outlier, blocked traffic volumes were modest and mostly involved expected infrastructure domains (Go module proxies, GitHub API, npm registry, telemetry endpoints) being denied due to allowlist gaps.
Note: Chart generation via the firewall-chart-generator sub-agent did not return usable output (no CHART1_URL/CHART2_URL or error field after two attempts), so trend charts are omitted from this report.
📊 Key Metrics
Metric
Value
Total network requests monitored
519,104
✅ Allowed
17,692
🚫 Blocked
501,412
Block rate
96.6%
Total unique blocked domains
23
Terminology Note: The 96.6% block rate is almost entirely driven by one workflow's repeated denied calls to registry.npmjs.org (500,250 of 501,412 blocked requests, ~99.8%). Excluding that outlier, the block rate across the remaining 260 runs is closer to 0.6% (1,162 blocked / 18,854 total non-outlier requests), which is a much more representative baseline for firewall health.
🚫 Top Blocked Domains
Domain
Times Blocked
Workflows
Category
registry.npmjs.org:443
500,250
Matt Pocock Skills Reviewer, PR Code Quality Reviewer, Tidy
Contribution Check, Daily Go Test Parallelizer, Daily Observability Report for AWF Firewall and MCP Gateway, Daily Syntax Error Quality Check, PR Sous Chef +2 more
Development Services
github.com:443
140
Contribution Check, Daily Go Test Parallelizer, Daily Malicious Code Scan Agent, Daily Rendering Scripts Verifier, Design Decision Gate 🏗️ +7 more
Development Services
ab.chatgpt.com:443
99
Daily Credit Limit Test, Daily Evals Feature Report, Daily Go Test Parallelizer, Daily Observability Report for AWF Firewall and MCP Gateway, Daily Storify +9 more
Total blocked requests: 500,243 (includes 4 blocked to other minor domains)
Total unique blocked domains: likely 1 dominant + minor
Most frequently blocked domain: registry.npmjs.org:443
Anomaly flag: this volume (avg. ~71,000 blocked requests/run) strongly suggests a retry storm or misconfigured install step repeatedly hitting the registry rather than distinct legitimate requests.
Workflow: Terminal Stylist (runs analyzed across period)
Domain
Blocked Count
Category
golang.org:443
13
Development Services
go.opentelemetry.io:443
5
Development Services
google.golang.org:443
5
Development Services
charm.land:443
4
Other
cloud.google.com:443
3
CDN
go.uber.org:443
2
Development Services
go.yaml.in:443
2
Development Services
gonum.org:443
1
Development Services
gopkg.in:443
1
Development Services
Total blocked requests: 370 (including proxy.golang.org contribution)
Most frequently blocked domain: proxy.golang.org:443 / golang.org:443
Likely cause: Go module dependency resolution reaching module proxy/source domains not on the allowlist.
Workflow: Daily Go Test Parallelizer (runs analyzed)
Domain
Blocked Count
Category
api.github.com:443
(part of 313 total)
Development Services
github.com:443
(part of 140 total)
Development Services
ab.chatgpt.com:443
(part of 99 total)
Analytics/Tracking
pypi.org:443
6
Development Services
raw.githubusercontent.com:443
4
Development Services
cafe.github.com:443
3
Other
Total blocked requests: 340
Most frequently blocked domain: api.github.com:443 / github.com:443
Likely cause: GitHub API/content calls during Go test setup not covered by the network allowlist.
Investigate the PR Code Quality Reviewer npm retry storm immediately. 500,239 blocked requests to registry.npmjs.org in 7 runs (avg. ~71,500/run) is not normal npm traffic — this is almost certainly a broken retry loop (e.g., an npm install/npx step retrying on every failure without backoff) that is generating firewall noise, wasting CI minutes, and could mask real issues. Recommend: add registry.npmjs.org to the allowlist for this workflow if npm access is legitimately needed, and separately fix the retry logic regardless.
Allowlist common Go tooling domains for Terminal Stylist and other Go-based workflows: proxy.golang.org, golang.org, go.opentelemetry.io, google.golang.org, go.uber.org, go.yaml.in, gonum.org, gopkg.in are all legitimate Go module ecosystem domains. Their appearance as blocked (with 0 allowed count) suggests the network allowlist for Go-based workflows needs to include the module proxy domain set.
Review GitHub API domain coverage (api.github.com, github.com, cafe.github.com, codeload.github.com, raw.githubusercontent.com) for workflows like Daily Go Test Parallelizer, Contribution Check, and Design Decision Gate. These are legitimate GitHub endpoints and should likely be allowlisted rather than blocked, especially since some also show partial "allowed" counts (e.g., github.com:443 had 6 allowed vs 140 blocked), indicating inconsistent/partial allowlist coverage rather than a clean deny.
ab.chatgpt.com and o205451.ingest.us.sentry.io appear across a very wide set of workflows (14 and 26 respectively) with mixed allow/block outcomes (e.g., Sentry had 696 allowed vs only 99 blocked). This pattern suggests these are shared telemetry/analytics endpoints baked into the base engine image — worth allowlisting centrally at the firewall policy level (rather than per-workflow) if the traffic is confirmed benign, to reduce noise across many unrelated workflows.
No suspicious/unknown external domains observed. All blocked domains map to recognizable developer tooling, GitHub infrastructure, or telemetry providers — no evidence of exfiltration attempts or unexpected third-party domains in this period.
Policy rule data unavailable. None of the 261 analyzed run summaries included policy_analysis (rule-level hit counts, SSL Bump/DLP status). If policy-level attribution is desired going forward, the firewall log capture pipeline should be updated to emit per-rule hit data into run_summary.json.
Report generated from 261 cached firewall-enabled workflow run summaries out of 611 total runs collected in the last 24 hours (350 runs had no firewall_analysis and were excluded; 0 summaries were skipped due to malformed/missing data).
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
Date: 2026-09-06 (report covers the last 24 hours of workflow activity)
This report analyzed 261 firewall-enabled workflow runs across 95 distinct workflows in the
github/gh-awrepository. Overall firewall activity was dominated by an extreme outlier: a single workflow (PR Code Quality Reviewer) accounted for over 500,000 blocked requests toregistry.npmjs.org:443across just 7 runs — this appears to be a retry/polling loop rather than normal traffic and skews the aggregate block rate heavily. Outside of that outlier, blocked traffic volumes were modest and mostly involved expected infrastructure domains (Go module proxies, GitHub API, npm registry, telemetry endpoints) being denied due to allowlist gaps.Note: Chart generation via the
firewall-chart-generatorsub-agent did not return usable output (no CHART1_URL/CHART2_URL or error field after two attempts), so trend charts are omitted from this report.📊 Key Metrics
🚫 Top Blocked Domains
(4a–4d Policy Rule Attribution omitted: no
policy_analysisdata was present in any of the 261 analyzed run summaries.)View Detailed Request Patterns by Workflow
Workflow: PR Code Quality Reviewer (7 runs analyzed)
Workflow: Terminal Stylist (runs analyzed across period)
Workflow: Daily Go Test Parallelizer (runs analyzed)
Workflow: Daily Syntax Error Quality Check (runs analyzed)
(Remaining workflows each contributed ≤26 blocked requests — see the complete domain list below for full attribution.)
View Complete Blocked Domains List
🛡️ Security Recommendations
Investigate the
PR Code Quality Reviewernpm retry storm immediately. 500,239 blocked requests toregistry.npmjs.orgin 7 runs (avg. ~71,500/run) is not normal npm traffic — this is almost certainly a broken retry loop (e.g., annpm install/npxstep retrying on every failure without backoff) that is generating firewall noise, wasting CI minutes, and could mask real issues. Recommend: addregistry.npmjs.orgto the allowlist for this workflow if npm access is legitimately needed, and separately fix the retry logic regardless.Allowlist common Go tooling domains for
Terminal Stylistand other Go-based workflows:proxy.golang.org,golang.org,go.opentelemetry.io,google.golang.org,go.uber.org,go.yaml.in,gonum.org,gopkg.inare all legitimate Go module ecosystem domains. Their appearance as blocked (with 0 allowed count) suggests the network allowlist for Go-based workflows needs to include the module proxy domain set.Review GitHub API domain coverage (
api.github.com,github.com,cafe.github.com,codeload.github.com,raw.githubusercontent.com) for workflows likeDaily Go Test Parallelizer,Contribution Check, andDesign Decision Gate. These are legitimate GitHub endpoints and should likely be allowlisted rather than blocked, especially since some also show partial "allowed" counts (e.g.,github.com:443had 6 allowed vs 140 blocked), indicating inconsistent/partial allowlist coverage rather than a clean deny.ab.chatgpt.comando205451.ingest.us.sentry.ioappear across a very wide set of workflows (14 and 26 respectively) with mixed allow/block outcomes (e.g., Sentry had 696 allowed vs only 99 blocked). This pattern suggests these are shared telemetry/analytics endpoints baked into the base engine image — worth allowlisting centrally at the firewall policy level (rather than per-workflow) if the traffic is confirmed benign, to reduce noise across many unrelated workflows.No suspicious/unknown external domains observed. All blocked domains map to recognizable developer tooling, GitHub infrastructure, or telemetry providers — no evidence of exfiltration attempts or unexpected third-party domains in this period.
Policy rule data unavailable. None of the 261 analyzed run summaries included
policy_analysis(rule-level hit counts, SSL Bump/DLP status). If policy-level attribution is desired going forward, the firewall log capture pipeline should be updated to emit per-rule hit data intorun_summary.json.Report generated from 261 cached firewall-enabled workflow run summaries out of 611 total runs collected in the last 24 hours (350 runs had no firewall_analysis and were excluded; 0 summaries were skipped due to malformed/missing data).
All reactions