[uk ai resilience] UK AI Governance: Recent-Changes Risk & Resilience Review (2026-09-02)

[github-actions[bot]]

Executive summary

Recent-changes-focused UK AI Open Code Risk & Resilience review of github/gh-aw over the last 7 days (since 2026-08-26). 257 commits merged, 55 flagged as security-signal commits, 41 open security-labeled issues, 10 open code-scanning alerts (0 secret-scanning alerts). No new critical exposure was found; the dominant pattern is remediation-velocity / alert-hygiene gaps: several CodeQL alerts already have open tracking issues asserting the underlying risk is a false positive or already fixed, yet the alerts remain undismissed in code scanning. One alert (README.md, #564) has no tracking issue at all and is the only genuinely untracked finding this run. Sub-agent dispatch (asset-tier-classifier, control-verifier, ai-risk-scorer) failed to return output on both attempts; the analysis below was synthesized directly from repo inspection and issue history with partial confidence as a result — recommend re-running sub-agents next cycle for deeper verification.

No repository-hiding or access-restriction action is recommended anywhere in this review; all findings are open-source/CI-visible artifacts, and outcomes favor faster alert lifecycle closure over concealment.

Asset graph summary (recent-change scoped)

Asset graph (7 areas touched by open alerts)

#	Area	Type	Ownership signal	Notes
1	pkg/parser/import_error.go	Go — parser error formatting	Unclear — no CODEOWNERS file exists in repo	New/untracked alert pair
2	pkg/workflow/graders_config.go	Go — workflow grader config	Unclear — no CODEOWNERS	Tracked in #57472, #54037 as recurring false positive
3	pkg/cli/add_package_manifest_includes.go	Go — CLI package manifest	Unclear — no CODEOWNERS	Tracked in #54037
4	scripts/ensure-docs-slide-pdf.js	Node — docs build pipeline	Unclear — no CODEOWNERS	Tracked in #53737
5	actionlint integration	Workflow tooling/binary	Unclear — no CODEOWNERS	Tracked in #57728, alert marked stale
6	pkg/cli/project_command.go	Go — GraphQL query construction	Unclear — no CODEOWNERS	Tracked in #52749, unresolved verification
7	README.md	Docs	Unclear — no CODEOWNERS	No tracking issue found

Scope note: the repository has no CODEOWNERS file, so ownership confidence is "Unclear" across every area — this is itself a structural finding (see Control Gaps below), not specific to any one file.

Tier classification table

Tier classification (A/B/C/D)

Area	Alert(s)	Existing issue	Tier	Rationale
pkg/parser/import_error.go	go/unsafe-quoting #671, #670	none found	B — Open With Conditions	Warning severity; needs a tracking issue and confirmation embedded data can't break quoting
pkg/workflow/graders_config.go	go/bad-redirect-check #669, #668	#57472, #54037 (Tier B)	B — Open With Conditions	Prior issues assert false positive; alert-dismissal hygiene gap, not live risk
pkg/cli/add_package_manifest_includes.go	go/bad-redirect-check #667	#54037 (Tier B)	B — Open With Conditions	Same pattern as above
scripts/ensure-docs-slide-pdf.js	js/http-to-file-access #663	#53737 (Tier C)	C — Restricted Pending Review	Untrusted network data written to file; docs pipeline, needs confirmed sanitization before closing
actionlint integration	workflow-out-of-context #653	#57728 (Tier C)	C — Restricted Pending Review	Issue claims binaries already remediated but alert undismissed — needs closure verification, not new code fix
pkg/cli/project_command.go	graphql-injection-sprintf #652, #651	#52749	B — Open With Conditions	Genuinely unresolved: needs parameterized-query verification
README.md	workflow-security-finding-1 #564	none	B — Open With Conditions	Only untracked finding; low severity (doc content) but needs an owner and issue

Control verification gaps

• Ownership controls: gap. No CODEOWNERS file exists in the repository root or .github/, so every changed area above has "unclear" ownership routing for security alerts.
• SDLC controls: present. CodeQL, actionlint, and JS scanning are actively running (evidenced by the 10 open alerts themselves and prior remediation issues) — detection pipeline is healthy.
• Dependency controls: no dependency-specific alerts in this window (secret-scanning empty; no Dependabot alerts surfaced in this batch).
• Secret exposure controls: no open secret-scanning alerts — clean for this window.
• Runtime observability: unable to confirm without deeper file-level review (sub-agent verification failed); flagged as a follow-up.
• Recovery/patch-readiness: gap. Three alerts ([Custom Engine Test] Test Issue Created by Custom Engine #669/Codex fails to report failure if unauthorised #668, [Custom Engine Test] Test Pull Request - Custom Engine Safe Output #667, #653) have paired issues ([uk-ai-resilience] UK AI Governance: recurring go/bad-redirect-check false positive across path-traversal guards (Tier B) #57472, [uk-ai-resilience] UK AI Governance: unreviewed CodeQL go/bad-redirect-check alert in add_package_manifest.go (Tier B) #54037, [uk-ai-resilience] Stale code-scanning alert #653 not dismissed despite remediated actionlint/gosec binaries (Tier C) #57728) asserting the finding is a false positive or already remediated in code, yet the alert remains open and undismissed in code scanning. This is a recurring alert-lifecycle/dismissal-hygiene gap that inflates the open-alert count without reflecting live risk, and reduces confidence in dashboards used for prioritization.

Risk-scoring table and rationale

AI-aware risk scoring

Area	Exposure Amplification	Patchability	Detectability	Operational Fragility	Ownership Confidence	Tier
pkg/parser/import_error.go	Low	High	Medium	Low	Low (no CODEOWNERS)	B
pkg/workflow/graders_config.go	Low	High (likely already fixed)	High	Low	Low	B
pkg/cli/add_package_manifest_includes.go	Low	High	High	Low	Low	B
scripts/ensure-docs-slide-pdf.js	Medium (writes untrusted network data to disk)	Medium	Medium	Medium	Low	C
actionlint integration	Low	High (claimed done)	High	Low (process gap only)	Low	C
pkg/cli/project_command.go	Medium (GraphQL query injection potential)	Medium (needs verification)	Medium	Medium	Low	B
README.md	Low	High	Low (no issue tracking it)	Low	Low (no owner, no issue)	B

Rationale: the three false-positive/stale cases (#57472/#54037, #57728) score low on exposure amplification and high on patchability — the real gap is process (un-dismissed alerts), not code risk, so remediation should focus on alert lifecycle closure rather than new code changes. project_command.go (#652/#651) and ensure-docs-slide-pdf.js (#663) carry genuine unresolved verification work and stay at moderate priority. README.md (#564) is elevated to B purely because it lacks any tracking issue — a governance-coverage gap independent of severity.

Remediation queue with SLAs

Priority	Area	Action	SLA
High	actionlint (#653/#57728), graders_config.go (#669/#668/#57472/#54037), add_package_manifest_includes.go (#667/#54037)	Verify remediation is actually merged/live, then dismiss the CodeQL alerts with a documented reason (not "won't fix" silently)	7 days
High	pkg/cli/project_command.go (#652/#651/#52749)	Confirm GraphQL query construction is parameterized/escaped; close or downgrade issue with evidence	14 days
Medium	scripts/ensure-docs-slide-pdf.js (#663/#53737)	Confirm sanitization of untrusted network content before file write; add regression test	14 days
Medium	pkg/parser/import_error.go (#671/#670)	Open a tracking issue; confirm quoted data can't embed unescaped quotes from import paths	14 days
Medium	README.md (#564)	Open a tracking issue and assign an owner; review flagged content	14 days
Low (governance)	Repository-wide	Add a CODEOWNERS file to give every security-relevant path a clear ownership/escalation route	30 days

Exception register

No temporary exceptions are proposed this run — all findings above are routed to remediation actions with SLAs rather than exception grants. If a future review needs to grant a temporary exception (e.g., for a confirmed-benign alert pending a batch dismissal), it must include: threat hypothesis, exploit acceleration claim, operational weakness, expiry date (≤30 days), and mitigation plan — none met that bar this cycle.

Operational metrics baseline

• MTTR proxy: 3 of 10 open alerts ([Custom Engine Test] Test Issue Created by Custom Engine #669/Codex fails to report failure if unauthorised #668, [Custom Engine Test] Test Pull Request - Custom Engine Safe Output #667, #653) reference remediation issues open for weeks without alert dismissal — indicates alert-closure MTTR is currently unbounded/untracked.
• Ownership coverage: 0% — no CODEOWNERS file covers any of the 7 flagged areas.
• Unsupported dependency ratio: not assessable this cycle (no dependency-specific alerts surfaced; sub-agent verification unavailable).
• Exception aging: N/A — no active exceptions on record for this window.
• Exposure without recovery capability: 1 of 7 areas (README.md, [Custom Engine Test] Test Issue Created by Custom Engine #564) has zero tracking/recovery path currently assigned.

Limitation note: the three specialist sub-agents (asset-tier-classifier, control-verifier, ai-risk-scorer) were dispatched twice each and returned no output both times; this report was synthesized directly from repo/issue evidence with partial confidence. Recommend re-running sub-agent dispatch next cycle once the underlying issue is resolved.

Generated by UK AI Operational Resilience · copilot · auto · 43.2 AIC · ⌖ 6.81 AIC · ⊞ 8.1K · ◷

• expires on Sep 5, 2026, 7:41 AM UTC-08:00