implementation-roadmap.md

April 3, 2026 · View on GitHub

Roadmap Overview

The framework rolls out across five phases, each introducing a new agent domain. Phases 1 and 2 leverage proven implementations. Phases 3 through 5 build on established patterns with new agent definitions. Cross-cutting centralized governance applies to all phases.

gantt
    title Agentic Accelerator Framework Rollout
    dateFormat  YYYY-MM-DD
    axisFormat  %b %Y

    section Phase 1: Security
    Security Agents (proven)           :done, p1, 2025-01-01, 2025-06-30
    Org-wide rollout via .github-private :done, p1r, 2025-07-01, 2025-09-30

    section Phase 2: Accessibility
    A11y Scanner + Agents (proven)     :done, p2, 2025-04-01, 2025-09-30
    Multi-site scanning rollout        :active, p2r, 2025-10-01, 2026-03-31

    section Phase 3: Code Quality
    Coverage-to-SARIF + Agents         :active, p3, 2026-01-01, 2026-06-30
    Quality gate enforcement           :active, p3r, 2026-04-01, 2026-09-30

    section Phase 4: FinOps
    Cost Analysis Agents               :active, p4, 2026-04-01, 2026-09-30
    Deployment cost gate               :active, p4r, 2026-07-01, 2026-12-31

    section Phase 5: Prompt Security
    APM audit + CI integration         :active, p5, 2026-06-01, 2026-09-30
    Supply chain monitoring            :p5r, 2026-10-01, 2027-03-31

    section Cross-Cutting
    Defender for Cloud + Dashboards    :crit, cc, 2025-01-01, 2027-03-31

Phase 1: Security Agents

AttributeDetail
StatusImplemented
Repositorydevopsabcs-engineering/.github-private
AgentsSecurityAgent, SecurityReviewerAgent, SecurityPlanCreator, PipelineSecurityAgent, IaCSecurityAgent, SupplyChainSecurityAgent
SARIF Categorysecurity/

Proof of Value

Demonstrated at the TT343 session ("Agentic AI for DevSecOps: Transforming Security with GHAS and GHCP") with a live demo using the gh-advsec-devsecops repository. Six custom agents produced actionable security reports covering OWASP Top 10, CIS Azure benchmarks, pipeline hardening, IaC misconfiguration, and supply chain risks.

Actions

  • Roll out organization-wide via .github-private
  • Enable in VS Code for pre-commit scanning
  • Configure CI SARIF upload to GitHub Code Scanning

Phase 2: Accessibility Agents

AttributeDetail
StatusImplemented
Repositoryaccessibility-scan-demo-app (scanner), .github-private (agents)
AgentsA11yDetector, A11yResolver
SARIF Categoryaccessibility-scan/

Proof of Value

A three-engine accessibility scanner (axe-core, IBM Equal Access, custom Playwright checks) produces SARIF output compliant with WCAG 2.2 Level AA. The scanner integrates with GitHub Actions for CI gating and uploads results to Code Scanning. Findings flow through to Defender for Cloud for centralized governance.

The Detector and Resolver agent pair demonstrates the handoff pattern: the Detector identifies violations using static and runtime analysis, hands off to the Resolver for automated remediation, and then re-scans for verification.

Actions

  • Onboard additional applications for accessibility scanning
  • Add scheduled production URL scans via cron workflows
  • Expand custom Playwright checks based on organizational patterns

Phase 3: Code Quality Agents

AttributeDetail
StatusActive — agents defined, coverage gate implemented, domain skill and scaffolding artifacts created
AgentsCodeQualityDetector, TestGenerator
SARIF Categorycode-quality/coverage/

Proof of Value

A coverage gate that fails the CI pipeline when code coverage drops below 80% for any file or function. Below-threshold functions are reported as SARIF results, appearing in GitHub Code Scanning alongside security and accessibility findings. The Test Generator agent reads uncovered functions and produces tests covering happy path and error paths.

Implementation Pattern

Follows the Detector and Resolver pattern from Accessibility:

  1. Code Quality Detector reads coverage reports (lcov, cobertura, JSON)
  2. Identifies files and functions below the 80% threshold
  3. Reports findings as SARIF with ruleId values: uncovered-function, uncovered-branch, coverage-threshold-violation
  4. Hands off to Test Generator for automated test creation
  5. Re-measures coverage after test generation

Multi-Language Coverage Tools

LanguageToolFormats
JavaScript/TypeScriptVitest, Jest, c8, Istanbullcov, cobertura, JSON
Pythoncoverage.py, pytest-covlcov, XML, JSON
.NET/C#dotnet test + coverletcobertura, opencover, lcov
JavaJaCoCo, CoberturaXML, HTML, CSV
Gogo test -coverprofileGo cover profile to lcov

Actions

  • Agent definitions (code-quality-detector.agent.md and test-generator.agent.md) are in place
  • Domain skill (skills/code-quality-scan/SKILL.md) documents the 4-tool architecture with MegaLinter orchestration
  • Scan and fix prompts (code-quality-scan.prompt.md, code-quality-fix.prompt.md) route to existing agents
  • Domain scaffolding system (DomainScaffolder agent, domain-scaffolding/SKILL.md, scaffold-domain.prompt.md) enables automated repo creation
  • GitHub Actions sample (samples/github-actions/code-quality-scan.yml) provides CI workflow template
  • Roll out organization-wide via .github-private using deploy-to-github-private.yml
  • Refine coverage-to-SARIF converter
  • Enforce quality gate workflow (code-quality.yml) in CI

Phase 4: FinOps / Cost Analysis Agents

AttributeDetail
StatusActive — agents defined, cost gate workflow implemented
AgentsCostAnalysisAgent, FinOpsGovernanceAgent, CostAnomalyDetector, CostOptimizerAgent, DeploymentCostGateAgent
SARIF Categoryfinops-finding/v1

Proof of Value

A deployment cost gate that blocks infrastructure changes exceeding the defined budget. The agent queries Azure Cost Management APIs, evaluates proposed IaC changes against budget constraints, and produces SARIF findings for budget overspend, cost anomalies, untagged resources, and optimization opportunities. Findings appear in Code Scanning alongside security alerts.

Azure Cost Management API Integration

APIPurpose
Cost Management QueryAd-hoc cost queries by tag, resource group, or service
BudgetsCreate and manage cost budgets with alerts
ExportsRecurring CSV to Azure Storage
Scheduled ActionsAnomaly detection alert rules
ForecastsProject future costs
Azure AdvisorOptimization recommendations

Actions

  • Five agent definitions are in place in agents/
  • Roll out organization-wide via .github-private using deploy-to-github-private.yml
  • Integrate with Azure Cost Management SDK
  • Enforce deployment cost gate via finops-cost-gate.yml workflow
  • Implement repository-to-cost attribution through ProjectName tagging

Phase 5: Prompt File Security

AttributeDetail
StatusActive — APM audit workflow implemented, CODEOWNERS protection in place
Key ToolAPM (Agent Package Manager) by Daniel Meppiel
SARIF Categoryagent-config-scan/

Proof of Value

APM's apm audit command detects hidden Unicode characters (zero-width spaces, bidi overrides, tag characters, Glassworm attack vectors) embedded in agent configuration files. The microsoft/apm-action GitHub Action integrates this scanning into CI/CD, blocking compromised agent configurations before they reach production. Combined with CODEOWNERS protection for agent config directories, this creates defense-in-depth for the AI supply chain.

APM Content Security Scanning

SeverityDetections
CriticalTag characters (U+E0001 to U+E007F), bidi overrides (U+202A to U+202E, U+2066 to U+2069), variation selectors 17 to 256 (Glassworm attack vector)
WarningZero-width spaces/joiners, variation selectors 1 to 15, bidi marks, invisible operators, annotation markers, deprecated formatting, soft hyphen, mid-file BOM
InfoNon-breaking spaces, unusual whitespace, emoji presentation selector

OWASP LLM Top 10 Alignment

OWASP RiskRelevance
LLM01 Prompt InjectionMalicious instructions in config files
LLM03 Supply ChainConfig files are part of the LLM supply chain
LLM06 Excessive AgencyConfig files define tool access and autonomy level
LLM07 System Prompt LeakageConfig files are system prompts; leaked configurations expose architecture

Actions

  • apm.yml manifest and apm-security.yml workflow are in place
  • .github/CODEOWNERS enforces security-team review for agent config paths
  • Run apm audit as a pre-commit check in VS Code
  • Expand supply chain monitoring coverage

Cross-Cutting: Centralized Governance

Centralized governance applies across all five phases and ensures findings from every domain flow into a unified view.

Components

ComponentRole
GitHub Security OverviewOrg-wide dashboard of all Code Scanning alerts, filtered by SARIF category
Microsoft Defender for CloudUnified security posture across GitHub, ADO, and runtime environments
Defender for DevOpsDevOps-specific findings console across GitHub, ADO, and GitLab
Power BI AdvSec ReportCompensates for ADO Security Overview API gap with star schema analytics
Security CampaignsBulk remediation of vulnerabilities with Copilot Autofix (GitHub-only)

Integration Requirements

  • GitHub Advanced Security (GHAS) enabled on all repositories
  • GitHub organization connected to Defender for Cloud via GitHub connector
  • ADO organization connected to Defender for Cloud via ADO connector (for hybrid environments)
  • Unified SARIF categories configured across all scan types (see the SARIF Integration Reference)
  • security-events: write permission in CI workflows

Governance Data Flow

graph LR
    subgraph "Scan Domains"
        S[Security Agents]
        A[Accessibility Agents]
        Q[Code Quality Agents]
        F[FinOps Agents]
        P[Prompt Security - APM]
    end

    subgraph "SARIF Upload"
        S --> SARIF[SARIF v2.1.0]
        A --> SARIF
        Q --> SARIF
        F --> SARIF
        P --> SARIF
    end

    subgraph "Centralized Governance"
        SARIF --> GH[GitHub Security Overview]
        SARIF --> ADO[ADO Advanced Security]
        GH --> MDC[Defender for Cloud]
        ADO --> MDC
        ADO --> PBI[Power BI AdvSec Report]
        MDC --> DFD[Defender for DevOps]
    end

References