[security-observability] Daily Security Observability Report — 2026-11-12

[github-actions[bot]]

Executive Summary

Coverage is partial. Startup found 88 eligible completed runs and selected 20, newest first. All 20 were audited successfully, and 68 were omitted by the cap. Every selected run was created on 2026-11-12 between 05:44 and 15:50 UTC. This report is therefore a same-day snapshot, not a seven-day census.

Firewall: 2,945 network requests were observed, and 2,944 (99.97%) were allowed. One request was blocked, in the Daily Fact run, and its domain was not resolved ((unknown)). No policy rule hits or denials were recorded across the 20 runs.

DIFC: 366 integrity-filtered MCP tool calls across 14 workflows, all on the github server. 365 cite integrity and 1 cites secrecy. Breaking Change Checker (131), Daily Assign Issue To User (92), and Daily Documentation Updater (88) account for 85% of events.

---

🔥 Firewall Analysis

Key Firewall Metrics

Metric	Value
Runs audited (selected, all with firewall data)	20 of 88 eligible
Total network requests monitored	2,945
✅ Allowed requests	2,944
🚫 Blocked requests	1
Block rate	0.03%
Unique blocked domains	1 ((unknown), unresolved)
Policy rule hits / denials	0 / 0 (20 runs, 9 to 11 rules each, SSL Bump and DLP disabled)

📈 Firewall Request Trends

This chart shows allowed and blocked requests per audited run. It is not a time series. Blocked traffic is negligible: one request in one run. Daily Documentation Updater made 1,924 requests, 65% of the total.

Top Blocked Domains

Only one blocked entry exists, and its domain was not resolved. It is too little signal to identify a concerning destination or an overly strict rule.

Domain	Times Blocked	Workflows	Category
(unknown)	1	Daily Fact	Unresolved

View Detailed Request Patterns by Workflow

Workflow	Run ID	Total	Allowed	Blocked
Daily Documentation Updater	43201995004	1,924	1,924	0
Daily A/B Testing Advisor	43200335556	292	292	0
Daily Assign Issue To User	43164100570	72	72	0
Agent Performance Analyzer - Meta-Orchestrator	43219879898	71	71	0
Issue Triage Agent	43239245914	54	54	0
Go Fan	43179043780	51	51	0
GitHub MCP Structural Analysis	43213194568	49	49	0
Design Decision Gate 🏗️	43252147681	45	45	0
Auto-Triage Issues	43217265117	45	45	0
Code Scanning Fixer	43171390896	38	38	0
Avenger	43227450902	37	37	0
Workflow Normalizer	43220217787	37	37	0
Avenger	43237361751	33	33	0
Ponytail Reviewer	43252148275	31	31	0
Breaking Change Checker	43241935147	31	31	0
PR Specification Reviewer	43252148046	30	30	0
Code Scanning Fixer	43222248452	30	30	0
Daily PR Code Quality Review	43163357763	28	28	0
Daily Fact	43235053587	25	24	1
Test Quality Sentinel	43252148149	22	22	0

Most allowed traffic goes to packagefeedproxy.microsoft.io:443 (1,795), api.githubcopilot.com:443 (428), and the Sentry and Grafana telemetry endpoints (266 each).

View Complete Blocked Domains List

• (unknown): 1 blocked request (Daily Fact, run 43235053587)

🔒 Firewall Security Recommendations

1. Check the unresolved block in Daily Fact (run 43235053587). The (unknown) label means the audit could not attribute the domain. Check the raw firewall log for that run before changing any allowlist.
2. No allowlist change is indicated by this window. The only block is a single unresolved request.
3. Expand coverage. The 68 omitted runs, and any runs from before 2026-11-12, were not audited. Re-run with a broader selection before drawing seven-day conclusions.

---

🔒 DIFC Integrity Analysis

Key DIFC Metrics

Metric	Value
Total filtered events	366
Unique tools filtered	10
Unique workflows affected	14
Unique author logins	55 (272 events have no author login)
Most common filter reason	Integrity (365 of 366 events)
Busiest day	2026-11-12 (366 events, all in the audited window)

📈 DIFC Events Over Time

All 366 events fall on one day, 2026-11-12, so the chart shows the audited selection, not a multi-day pattern. Hourly volume peaks at 14:00 UTC (154 events), followed by 05:00 (93) and 10:00 (90).

🔧 Top Filtered Tools

list_commits (132) and list_issues (93) account for 61% of events. search_pull_requests (44), pull_request_read (40), and search_issues (29) make up most of the rest. Breaking Change Checker drives list_commits (64), pull_request_read (39), and search_pull_requests (28). Daily Assign Issue To User accounts for 92 of the 93 list_issues events.

🏷️ Filter Reasons and Tags

Integrity filtering dominates: 365 events cite a "lower integrity than agent requires" reason (the approved threshold). Integrity tags were none:public (366 events, all of them) and unapproved:public (90). Three events carry the private secrecy tag, but only one has a secrecy-class reason: a list_code_scanning_alerts call in GitHub MCP Structural Analysis. The other two are get_code_scanning_alert calls in Code Scanning Fixer, which cite integrity on a security-alert resource.

📋 Per-Workflow DIFC Breakdown

Workflow	Filtered Events
Breaking Change Checker	131
Daily Assign Issue To User	92
Daily Documentation Updater	88
Daily Fact	20
Go Fan	13
Agent Performance Analyzer - Meta-Orchestrator	7
Avenger	4
GitHub MCP Structural Analysis	3
Code Scanning Fixer	2
Daily A/B Testing Advisor	2
Issue Triage Agent	1
Workflow Normalizer	1
Auto-Triage Issues	1
Daily PR Code Quality Review	1

📋 Per-Server DIFC Breakdown

MCP Server	Filtered Events
github	366

👤 Per-User DIFC Breakdown

Author Login	Filtered Events
(unknown)	272
dsyme	12
lpcox	7
yskopets	6
jonas-harmoney	6
bbonafed	3
akitana-airtanker	3
timjansenortec	2
benissimo	2
theletterf	2
pelikhan	2
v1v	2

The remaining 44 logins account for 47 events: 41 logins have one event each and 3 have two.

💡 DIFC Tuning Recommendations

1. Breaking Change Checker (131 events). Most of these are list_commits, pull_request_read, and search_pull_requests reads. Check whether the analysis needs commits and PRs from unapproved authors, or whether its queries can be narrowed to approved content.
2. Daily Assign Issue To User (92 events, all list_issues) and Daily Documentation Updater (88 events). Review whether these workflows need unscoped issue or commit lists. Scoping the queries reduces filtered reads. If the content is needed, record the reason for approving the relevant authors.
3. Author attribution gap (272 events). Most events have no author login, so per-user data is incomplete. Fix author extraction in the audit before using per-user data for policy decisions.
4. Coverage. Re-run with a seven-day selection. The current set covers one day, so the timeline and tool counts are not representative.

---

Data Notes

• Selection: 20 of 88 eligible runs audited (partial). 68 omitted, 0 failed, 2 concurrent audits. Every selected run's output was parsed successfully.
• DIFC events come from mcp_tool_usage.filtered_events. The gateway_analysis.filtered_events field named in the workflow spec is absent from the audit output.
• Firewall chart 1 is per-run, not a 30-day time series, because the audited runs span one day.
• Of the 8 firewall-enabled runs listed at startup, 4 are in the audited set.

Generated by the Daily Security Observability workflow (consolidated from Daily Firewall Reporter + Daily DIFC Analyzer)
Analysis window: runs created 2026-11-12 (partial, see coverage note) | Repository: github/gh-aw
Run: https://github.com/github/gh-aw/actions/runs/43252632377

Generated by 🔒 Daily Security Observability Report · copilot · routed: haiku55 xhigh · ~8 AIC · ⌖ ~2 AIC · ⊞ 12.3K · ◷

• expires on Nov 15, 2026, 8:20 AM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Daily Security Observability Report.

A newer discussion is available at Discussion #77509.