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 run audited 20 of 100 eligible runs from the 7-day window (partial coverage; 80 runs were omitted by the 20-audit cap). All 20 audits succeeded. Firewall traffic was almost entirely allowed: 3,117 of 3,118 requests (99.97%) passed, and the single block was an unresolved domain from Daily Fact.
DIFC integrity filtering recorded 2,963 events across 18 runs, but 82% (2,439) come from one failed run, Copilot PR Conversation NLP Analysis (43407559464). That run logged 1,609 MCP tool calls in a single turn and ended with report_incomplete. Excluding it, 524 events remain, led by Dependabot Dependency Checker (234) and Workflow Normalizer (100). The same workflow had clean network traffic (50 requests, 0 blocked), so its filtering is an MCP integrity issue, not a network one.
🔥 Firewall Analysis
Coverage
Metric
Value
Eligible runs (7-day window)
100
Selected and audited
20 (20 succeeded, 0 failed)
Omitted by audit cap
80
Audited runs with firewall data
18
Audited runs without firewall data
2 (PR Sous Chef: 43456621340, 43455061983)
Key Firewall Metrics
Metric
Value
Workflows analyzed (firewall-enabled)
13 (across 18 runs)
Total network requests monitored
3,118
✅ Allowed requests
3,117
🚫 Blocked requests
1
Block rate
0.03%
Unique blocked domains
1 (unresolved (unknown))
📈 Firewall Request Trends
Chart shows requests per audited run, not a 30-day daily series, because all selected runs fall on 2026-11-13. Blocked traffic is negligible.
Top Blocked Domains
Only one blocked request was observed, to an unresolved (unknown) domain from Daily Fact. No allowlist change is indicated.
Most Frequently Blocked Domains
Domain
Times Blocked
Workflows
Category
(unknown)
1
Daily Fact
Unresolved
Policy Rule Attribution
📋 Policy: 10 rules (9 in one run), SSL Bump disabled, DLP disabled. Rule hits were empty in all 18 runs, and policy counters report 0 requests against 3,118 observed requests. This is an attribution gap, not evidence of zero traffic.
No allowlist changes are indicated by this window.
Check the unresolved (unknown) blocked request from Daily Fact to confirm it is not a misrouted call.
Fix policy rule attribution: rule hits are empty despite observed traffic, so rule-level data cannot be relied on yet.
🔒 DIFC Integrity Analysis
Key DIFC Metrics
Metric
Value
Total filtered events
2,963
Unique tools filtered
11
Unique workflows affected
13
Most common filter reason
get_review_comments with unresolved ref github/gh-aw#unknown (1,281 events)
Busiest day
2026-11-13 (2,963 events)
📈 DIFC Events Over Time
All events fall on a single day (2026-11-13), so this is not a 7-day trend. The spike is a single failed run (2,439 events between 10:21 and 10:37 UTC).
🔧 Top Filtered Tools
pull_request_read accounts for 2,458 events (83%), followed by search_issues (265) and list_issues (183).
🏷️ Filter Reasons and Tags
There are 761 distinct reason strings. Integrity tags: none:public (2,963), unapproved:public (1,444), approved:public (7), merged:public (6). Secrecy tag private appears on 43 events.
📋 Per-Workflow DIFC Breakdown
Workflow
Filtered Events
Copilot PR Conversation NLP Analysis
2,439
Dependabot Dependency Checker
234
Workflow Normalizer
100
Daily CLI Performance Agent
79
Issue Triage Agent
52
Daily Documentation Updater
33
Avenger
11
Daily Ecosystem Explorer
6
GitHub MCP Structural Analysis
3
Code Scanning Fixer
2
Go Fan
2
Daily Fact
1
Daily A/B Testing Advisor
1
📋 Per-Server DIFC Breakdown
MCP Server
Filtered Events
github
2,963
👤 Per-User DIFC Breakdown
Author Login
Filtered Events
unknown (no author attached)
1,752
copilot-pull-request-reviewer[bot]
906
github-advanced-security[bot]
246
lpcox
16
dsyme
10
pelikhan
6
Ahamed-1G
6
Others (14 users, 1–3 each)
21
💡 DIFC Tuning Recommendations
Investigate run 43407559464 before changing policy. It failed, made 1,609 tool calls in one turn, and drives 82% of events. Find out why it issued so many review-comment reads before deciding on any rule change.
Check resource labeling for review comments. 1,281 events reference github/gh-aw#unknown, so the PR number may not be resolved. Confirm this before treating these as real integrity violations.
Decide on bot-authored review content.copilot-pull-request-reviewer[bot] produced 906 unapproved:public events. Either treat that content as approved for the NLP analysis workflow or scope its reads so it does not fetch review threads.
Confirm private-secrecy reads are intended. 43 events carry a private secrecy tag, from Dependabot Dependency Checker (34), Daily Ecosystem Explorer (6), Code Scanning Fixer (2), and GitHub MCP Structural Analysis (1).
Methodology and Data Notes
DIFC events were read from mcp_tool_usage.filtered_events. The audit schema does not expose a gateway_analysis field.
Audit output for run 43407559464 exceeded the inline MCP limit and was read from its spilled payload file. The 2,439-event spike depends on that read.
Coverage is partial (20 of 100 eligible runs). Figures describe the audited subset, not a full seven-day census.
Generated by the Daily Security Observability workflow (consolidated from Daily Firewall Reporter + Daily DIFC Analyzer) Analysis window: Last 7 days (partial coverage: 20 of 100 eligible runs) | Repository: github/gh-aw Run: https://github.com/github/gh-aw/actions/runs/43457088125
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 run audited 20 of 100 eligible runs from the 7-day window (partial coverage; 80 runs were omitted by the 20-audit cap). All 20 audits succeeded. Firewall traffic was almost entirely allowed: 3,117 of 3,118 requests (99.97%) passed, and the single block was an unresolved domain from Daily Fact.
DIFC integrity filtering recorded 2,963 events across 18 runs, but 82% (2,439) come from one failed run, Copilot PR Conversation NLP Analysis (43407559464). That run logged 1,609 MCP tool calls in a single turn and ended with
report_incomplete. Excluding it, 524 events remain, led by Dependabot Dependency Checker (234) and Workflow Normalizer (100). The same workflow had clean network traffic (50 requests, 0 blocked), so its filtering is an MCP integrity issue, not a network one.🔥 Firewall Analysis
Coverage
Key Firewall Metrics
(unknown))📈 Firewall Request Trends
Chart shows requests per audited run, not a 30-day daily series, because all selected runs fall on 2026-11-13. Blocked traffic is negligible.
Top Blocked Domains
Only one blocked request was observed, to an unresolved
(unknown)domain from Daily Fact. No allowlist change is indicated.Most Frequently Blocked Domains
Policy Rule Attribution
📋 Policy: 10 rules (9 in one run), SSL Bump disabled, DLP disabled. Rule hits were empty in all 18 runs, and policy counters report 0 requests against 3,118 observed requests. This is an attribution gap, not evidence of zero traffic.
View Detailed Request Patterns by Workflow
Allowed domains observed: api.github.com, api.githubcopilot.com, api.openai.com, github.com, Sentry ingest, and Grafana OTLP gateway.
🔒 Firewall Security Recommendations
(unknown)blocked request from Daily Fact to confirm it is not a misrouted call.🔒 DIFC Integrity Analysis
Key DIFC Metrics
get_review_commentswith unresolved refgithub/gh-aw#unknown(1,281 events)📈 DIFC Events Over Time
All events fall on a single day (2026-11-13), so this is not a 7-day trend. The spike is a single failed run (2,439 events between 10:21 and 10:37 UTC).
🔧 Top Filtered Tools
pull_request_readaccounts for 2,458 events (83%), followed bysearch_issues(265) andlist_issues(183).🏷️ Filter Reasons and Tags
There are 761 distinct reason strings. Integrity tags:
none:public(2,963),unapproved:public(1,444),approved:public(7),merged:public(6). Secrecy tagprivateappears on 43 events.📋 Per-Workflow DIFC Breakdown
📋 Per-Server DIFC Breakdown
👤 Per-User DIFC Breakdown
💡 DIFC Tuning Recommendations
github/gh-aw#unknown, so the PR number may not be resolved. Confirm this before treating these as real integrity violations.copilot-pull-request-reviewer[bot]produced 906unapproved:publicevents. Either treat that content as approved for the NLP analysis workflow or scope its reads so it does not fetch review threads.privatesecrecy tag, from Dependabot Dependency Checker (34), Daily Ecosystem Explorer (6), Code Scanning Fixer (2), and GitHub MCP Structural Analysis (1).Methodology and Data Notes
mcp_tool_usage.filtered_events. The audit schema does not expose agateway_analysisfield.Generated by the Daily Security Observability workflow (consolidated from Daily Firewall Reporter + Daily DIFC Analyzer)
Analysis window: Last 7 days (partial coverage: 20 of 100 eligible runs) | Repository: github/gh-aw
Run: https://github.com/github/gh-aw/actions/runs/43457088125
All reactions