People & Professional Services Templates
Three templates for businesses whose product is people: billable delivery, recruiting, and bench management.
create-leafmesh my-project --template services-delivery # or:
create-leafmesh my-project --template recruiting-ops
create-leafmesh my-project --template bench-managementservices-delivery — Billable Engagement Work
18 agents · 2 crons — the sharpest dual case: billable people are both your biggest cost and all your revenue — and now the largest template in the catalog.
| Agent | Type | Role |
|---|---|---|
brief_intake_agent | LLM | Engagement brief → canonical schema |
research_agent ↔ context_agent | LLM | Bidirectional pair: prior-art and industry data ↔ client history and relationships |
consultant_agent | LLM | Consultant — reasons over research + context, then writes the first-pass deliverable with citations |
qa_agent | LLM | Citations, uncited claims, consistency, tone — can kick back to the consultant |
citation_agent | Programmatic | Every citation must resolve to a retrievable source |
delivery_router_agent | Programmatic | Routes the human editing pass: associate / senior / SME by stakes |
engagement_controller_agent | LLM | Engagement controller — billable-hours ledger plus statement-of-work-vs-actuals judgment → scope creep and change-order candidates |
engagement_status_drafter_agent | LLM | Weekly cron — client status report drafted from session activity |
status_sweep_orchestrator_agent | Programmatic | Weekly sweep — blocked and overdue engagements |
performance_review_agent | LLM | Consultant reviews weighted by client satisfaction + repeat business + outcomes |
doc_extraction_agent | LLM | At engagement close: mines deliverables for reusable IP, anonymises the client, tags reusability → firm knowledge base |
associate / senior_consultant / sme / engagement_lead / partner | Human | Five-tier delivery roster; the engagement lead always reviews before the client sees anything |
delivery_audit_logger_agent | External | Immutable log of every QA outcome, edit, approval |
How it works. Brief fans to research + context (which call each other), the consultant reasons over both and writes the draft → QA → citation check → routed to the right consultant tier for the human pass → engagement-lead gate → client. Citation-or-nothing is enforced mechanically. Around the core: the engagement controller's hours ledger feeds its own scope-creep judgment so change orders surface before the budget blows, weekly status reports write themselves, and closed engagements feed the knowledge base the next engagement draws on.
Value. In professional services, every hour saved is margin AND capacity simultaneously. The mesh does the routine share — research assembly, first drafts, status reports — while the citation gate and lead review keep quality client-grade. The IP-capture loop means the firm gets smarter with every engagement instead of re-learning per-project.
recruiting-ops — Source → Offer
13 agents · 2 crons
| Agent | Type | Role |
|---|---|---|
req_intake_agent | LLM | Requisition normalised; never invents salary bands |
sourcing_agent | Programmatic | LinkedIn / job boards / internal database |
resume_screener_agent | LLM | Filters against the requisition; bias-signal self-detection → HRBP |
hiring_assistant_agent | LLM | First-round async screen — 5-dimension weighted rubric → advance / borderline / decline |
interview_coordinator_agent | LLM | Interview coordinator — designs the loop (dimension coverage + diversity-aware panel + duration cap) and books the panel |
correspondence_drafter_agent | LLM | Candidate correspondence writer — personalised outreach, band-enforced offer letters (outside-band → HRBP), and bridge-preserving rejections in one role |
bias_audit_agent | Programmatic | Weekly cron — EEOC 4/5ths-rule disparate-impact audit across protected classes |
recruiting_ops_agent | Programmatic | Recruiting-ops analyst — daily thin-funnel sweep plus the stage-transition / drop-off / time-to-hire ledger |
recruiter / hiring_manager / recruiting_manager / hrbp | Human | HRBP holds bias, salary-band, and policy authority |
recruiting_audit_logger_agent | External | Immutable audit |
How it works. Requisition → sourcing → screening → async screen interview (rubric-scored, not vibes) → the coordinator designs and books the loop → offer. Bias protection runs at three layers: per-decision self-flags in screening, per-transcript keyword gates in the screen interview, and the weekly statistical audit that catches what individual checks can't — aggregate disparate impact.
Value. Recruiting is high-volume, deadline-driven, and legally exposed. The mesh compresses time-to-hire on the routine funnel while making fairness measurable — the 4/5ths audit produces the evidence that disparate-impact questions demand, before anyone asks for it.
bench-management — Consulting Bench & Staffing
11 agents · 1 cron
| Agent | Type | Role |
|---|---|---|
workforce_analytics_agent | Programmatic | Workforce analyst — 30/60/90-day capacity forecast with bench-ceiling alerts, the weekly per-consultant utilisation ledger, and the searchable skills inventory |
staffing_coordinator_agent | LLM | Staffing coordinator — matches people to opportunities by skills, then builds the assignment plan; conflicts → engagement lead |
talent_development_agent | LLM | Talent development partner — flags long-idle staff (2 weeks → reskilling bet, 4 weeks → practice lead), and drafts 6-month career paths: internal projects + certifications |
performance_review_agent | LLM | Reviews from engagement feedback + utilisation + skill growth |
esat_survey_agent | LLM | Anonymised pulse surveys; engagement index + theme extraction |
weekly_bench_sweep_agent | Programmatic | Weekly review cron |
staffing_partner / engagement_lead / talent_ops / practice_lead | Human | Assignment ownership → conflict resolution → reskilling → bench-ceiling authority |
bench_audit_logger_agent | External | Immutable audit |
How it works. The forecast → match → allocate chain runs on every new project and every weekly sweep — analytics feeding the staffing coordinator. Idle time triggers the talent-development agent's graduated response: 2 weeks → reskilling recommendation, 4 weeks → a career-path conversation with the practice lead. Utilisation, skills, and morale are all ledgered, so staffing decisions run on data instead of hallway knowledge.
Value. Bench time is pure cost; mis-staffing is pure churn risk. A 2-point utilisation improvement at a 100-consultant firm is roughly seven figures. The mesh closes the loop between who's free, what's coming, and what skills the market is asking for — and the anonymised pulse keeps you ahead of the resignations that idle benches breed.
Next: Templates Overview · Cloud Deploy