Engineering & Infrastructure

Engineering & Infrastructure Templates

Two templates for the technical organisation: the software delivery lifecycle and cloud cost operations.

create-leafmesh my-project --template engineering-sdlc # or: create-leafmesh my-project --template infra-cost-ops

engineering-sdlc — Ticket → Merge

14 agents · 2 crons

AgentTypeRole
spec_agentLLMTicket → testable spec + risk assessment; high-risk or public-API work forces architect review
junior_engineer_agentLLMImplements the spec; opens a draft PR
test_engineer_agentLLMWrites and runs tests (parallel with security)
security_scanner_agentProgrammaticVulnerability + dependency scan on every build
code_reviewer_agentLLMFan-in over build + test + security; kicks back or approves; HIGH security findings hard-block
regression_detector_agentProgrammaticTest flips and flakes from CI history; flips block merges
refactor_agentLLMWeekly tech-debt pass
oncall_alert_triager_agentLLMPagerDuty/Datadog/Sentry alerts → classified, right runbook attached
release_notes_drafter_agentLLMMerged PRs → categorised notes, security changes flagged
regulatory_monitor_agentLLMWeekly cron — PCI / HIPAA / FedRAMP / SOC 2 / ISO / CCPA / EU-AI-Act changes mapped to your controls
senior_engineer / architect / eng_managerHumanRisk-tiered review: senior for cross-cutting, architect for API/architecture, manager for merge
eng_audit_logger_agentExternalImmutable log of merges, approvals, halts

How it works. Spec → build → test and security in parallel → the reviewer fans in and is the only gate to merge approval. Risky specs detour through the architect — whose approval feeds back into the build loop, not past it. Around the core loop: flake detection guards CI trust, on-call triage compresses incident response, and the compliance watcher turns regulatory changes into engineering tickets before audit season.

Value. The routine share of engineering — boilerplate implementation, test writing, first-pass review, release notes — stops consuming senior engineers. They re-enter exactly where their judgment matters: security flags, API changes, architecture. Same team, more merged.


infra-cost-ops — Cloud Spend Tuning

9 agents · 2 crons

AgentTypeRole
cost_anomaly_agentLLMCost anomaly engineer — spike / drift / idle / new-resource classification with dollar impact *and* the right-sizing tune proposal, data-tier detection included
regulatory_monitor_agentProgrammaticCompliance veto: blocks region or encryption tunes that violate GDPR/DPDP/PIPL residency or PCI/HIPAA/FedRAMP encryption floors
cost_tuner_agentProgrammaticApplies approved tunes via cloud APIs
commitments_agentLLMCommitments analyst — weekly cron covering RI/SP utilisation, rebalance opportunities, expiry alerts, and reserved-capacity savings forecasts
iac_drift_agentProgrammaticTerraform/Pulumi intent vs actual cloud state
finops_governance_agentProgrammaticFinOps governance — tag-compliance ledger plus daily spend by cost centre, month-end forecast, overage alerts
sre_human / cto_humanHumanSRE on risky tunes and critical drift; CTO on data-tier and major cost decisions
infra_audit_logger_agentExternalImmutable log of every applied tune

How it works. The anomaly engineer classifies the cost event and proposes the right-sizing tune in one pass. Safe tunes auto-apply; data-tier changes always go to the SRE; and the compliance veto sits in front of the tuner — a tune that would move EU customer data out of region gets blocked before it's ever proposed to a human. Weekly commitment reviews keep reservation coverage honest; drift detection keeps infrastructure-as-code truthful.

Value. Cloud bills grow by default — idle resources, oversized instances, expired commitments nobody re-bought. The mesh applies the safe majority of savings automatically and routes the risky remainder with full context. The compliance veto means cost optimisation can never become a residency violation.


Next: Finance & Procurement · Templates Overview

Last updated: 11/10/2026