Semper Admin AI Field Manual
The Semper Admin AI Field Manual v3.0. 100 prompts across 10 categories. 53 pages. The operational AI prompt standard for United States Marine Corps administration. Built by Marines for Marines under the Semper Admin project at github.com/semperadmin. Five pillars: Accuracy over Speed, Verification before Submission, Accountability Stays with the Marine, Designed for the Fight, Open and Improving. CUI ceiling. Use only on DoD-authorized secured LLM systems with current ATO. Per NAVMC 5239.1, MARADMIN 018/26, EO 13556, and 32 CFR Part 2002.
Start Here If You Are New to the Field Manual
The SEMPER ADMIN AI FIELD MANUAL is the OPERATIONAL AI PROMPT STANDARD for United States Marine Corps administration. Version 3.0 ships 100 PROMPTS across 10 CATEGORIES in 53 pages. Built by Marines for Marines. CUI CEILING. Use only on DoD-authorized secured LLM systems with a current ATO. The manual establishes five operating pillars. ACCURACY OVER SPEED, VERIFICATION BEFORE SUBMISSION, ACCOUNTABILITY STAYS WITH THE MARINE, DESIGNED FOR THE FIGHT, and OPEN AND IMPROVING. The 10 categories cover Command Correspondence, Personnel Administration, Training Administration, Awards and Evaluations, Legal and Disciplinary, Logistics and Supply, Operational Planning, Financial Administration, Records and Information Management, and Leadership and Decision Making. The manual aligns to MARADMIN 018/26 (GenAI.mil platform), MARADMIN 214/26 (Mandatory Basic AI Course), MARADMIN 635/24 (Ethical Use), NAVMC 5239.1, EO 13556, 32 CFR Part 2002, and MCO 5231.4. Download the PDF below or visit github.com/semperadmin for the latest version.
Download
- Download the Field Manual PDF
- Visit github.com/semperadmin for the latest version and to contribute
The Five Pillars (Semper Admin Standard)
Accuracy Over Speed
A fast wrong answer is worse than a slow right one. Every output is a draft until a Marine verifies it. The model assists. The Marine decides. Speed is a byproduct of a system done right, not a goal pursued at the expense of accuracy.
Verification Before Submission
No AI output routes through the chain of command without human verification. Check names, dates, ranks, and regulatory citations against source systems. MCTFS, MOL, and the MCO are ground truth. The model is not.
Accountability Stays with the Marine
The Marine who submits the document owns it. AI assistance does not transfer accountability to the model, the developer, or the system. NAVMC 5239.1 is clear. You remain fully responsible for any AI-assisted document you sign or route.
Designed for the Fight
The library is built for S-1 shops, admin clerks, commanders, and SNCOs under time pressure in real units. Every prompt targets a real task. Every format matches a real requirement. Nothing is theoretical.
Open and Improving
A living system. Prompts are versioned. Regulations change. Report errors through your chain of command or at github.com/semperadmin. Accuracy is a team effort across the Total Force.
The 10 Prompt Categories
Each category contains 10 prompts. Classification tags are UNCLAS, FOUO, and CUI. CUI prompts require a DoD-authorized secured LLM system.
Category 01. Command Correspondence
Endorsements, SOPs, policies, MOUs, AAR, congressional inquiries, IG responses, complaint responses, policy revisions, information papers. Reference MCO 5216.20. 10 prompts.
Category 02. Personnel Administration
PCS/PCA orders, EAS separation packages, Page 11 entries, HUMS PTAD vs transfer, promotion warrant review, hardship discharge, legal hold, AMHS naval message, AWOL/desertion actions, HUMS personal request. Reference MCO 1070.12K. 10 prompts. See also wiki content in Life Events Overview, Reenlistment Eligibility and Process, and Separations and Retirement Overview.
Category 03. Training Administration
Training plans, T&R event briefs, MCTIMS documentation, deficiency reports, individual training record audits, readiness narratives, pre-deployment plans, annual training plans, safety risk assessments, training AARs. Reference MCO 1553.3B. 10 prompts.
Category 04. Awards and Evaluations
NAVCOMM, NAM, Meritorious Mast (NAVMC 10935), FITREP Sections B/C/I, adverse FITREP narratives, FITREP Section I prohibited comments scanner, CONDUS narrative, award package pre-submission review, MSM citation, promotion recommendation letter. Reference MCO 1610.7 and MCO 1650.19J. 10 prompts. Aligns with the wiki Awards parent group and Performance Evaluation parent group.
Category 05. Legal and Disciplinary
NJP preliminary inquiry, NJP notification and rights advisement, NJP punishment worksheet, admin sep notification procedure, admin sep board referral, Page 6 counseling statements, no contact orders, non-punitive LOIs, IG complaint intake, evidence inventory and chain of custody. Reference MCM 2024 and MCO 5800.16. 10 prompts. Aligns with the wiki Legal Services parent group.
Category 06. Logistics and Supply
Property accountability reconciliation, equipment readiness narratives, supply request justification, change of accountable officer, maintenance work order justification, GCSS-MC discrepancy reports, sensitive item inventory, deployment equipment prep, excess property disposition, Report of Survey packages. Reference MCO 4400.177G. 10 prompts.
Category 07. Operational Planning
OPORD frameworks, commander's intent statements, decision briefs, FRAGORDs, intelligence preparation summaries, risk assessments, staff estimates, WARNORDs, execution checklists, CCIRs. Reference MCWP 5-10 and MCDP 1. 10 prompts.
Category 08. Financial Administration
Travel voucher checklists, pay discrepancy resolution, budget justification, debt waivers, special pay authorization, GTC audits, advance pay, financial readiness counseling, BAH entitlement determinations, travel claim discrepancy letters. Reference JTR, DoDFMR, DTS. 10 prompts. Aligns with the wiki Pay and Entitlements parent group, DTS parent group, and GTCC parent group.
Category 09. Records and Information Management
FOIA responses, Privacy Act responses, records retention schedules, CUI marking guides, records disposition requests, information system authorization support, data call response, PII breach reports, records legal hold notices, command information program content plans. Reference MCO 5210.11F and 32 CFR Part 2002. 10 prompts.
Category 10. Leadership and Decision Making
Commander's assessment framework, ethical decision-making framework, climate assessment survey design, leader development plans, unit problem-solving briefs, succession planning, AAR facilitation guide, command philosophy development, difficult conversation preparation, unit standard setting framework. Reference MCDP 1 and MCWP 6-10. 10 prompts.
The 20 Laws of LLM Prompts
The Field Manual codifies 20 laws across four phases. Every prompt is a direct order. Write it like one.
Phase 1. Before the Prompt (Foundation)
- Law 1. Be Direct and Clear.
- Law 2. Provide Context First.
- Law 3. Ground in Real Data.
- Law 4. Set Boundaries.
- Law 5. Make It Personal.
- Law 6. Include What You've Tried.
Phase 2. Structure the Prompt
- Law 7. Ask for Edge Cases and Trade-offs.
- Law 8. Use Artifacts Wisely.
- Law 9. Control the Format.
- Law 10. Lock in the Persona.
- Law 11. Guide the Tone.
- Law 12. Avoid Prompt Conflicts.
Phase 3. After the Response
- Law 13. Challenge and Refine.
- Law 14. Iterate in Layers.
- Law 15. Invite Pushback.
- Law 16. Verify Facts.
Phase 4. System Thinking
- Law 17. Benchmark Against the Best.
- Law 18. Use Constraints as Creative Force.
- Law 19. Separate Thinking vs Final Answer.
- Law 20. Manage Context.
USMC Application
- Apply all 20 laws every time you use this library.
- Laws 3, 9, 16 are non-negotiable for CUI prompts.
- Laws 14 and 19 cut hallucination risk by 60 percent or more.
- Law 10 (Persona) plus Law 4 (Boundaries) yield highest-quality output.
- Law 6 (Include What You Tried) saves the most rework.
CUI Handling (Three Tiers)
The Field Manual enforces a clear classification framework per EO 13556, 32 CFR Part 2002, and MCO 5510.20.
UNCLAS
- System required. Any approved LLM system.
- Example content. Command correspondence, SOPs, award citations, training plans.
- Handling controls. None required beyond standard session hygiene.
FOUO
- System required. Access-controlled approved system. No public cloud without authorization.
- Example content. Personnel actions, IG responses, readiness reports, budget justifications.
- Handling controls. Do not email without encryption. Do not print on uncontrolled printers.
CUI
- System required. DoD-authorized secured LLM ONLY. Verify ATO before use.
- Example content. NJP packages, separation actions, pay records, medical, legal proceedings.
- Handling controls. Banner mark all outputs. Destroy per CUI Registry. Log all access.
Non-Negotiable Rules
- System Authorization. Verify your LLM system holds an ATO for CUI before processing any CUI-tagged prompt.
- No Classified Input. No prompt in the library handles SECRET or above. If your input is classified, stop.
- PII in Placeholders. PII enters prompts only through [PLACEHOLDER] variables filled at runtime.
- Mark All Outputs. Any output from a CUI or FOUO prompt must be banner-marked before saving or transmitting.
- Destruction Required. CUI outputs must be destroyed per CUI Registry requirements. Log the destruction.
Trust but Verify (Mandatory Verification Checklist)
Every output gets verified before routing. The checklist covers.
Identity
- Marine full name spelled correctly and matches MOL.
- Rank correct and matches current grade in MCTFS.
- MOS and PMOS verified against service record.
- Billet title matches current assignment.
Dates
- All dates formatted correctly per MCO or JTR requirement.
- Reporting period dates match FITREP data.
- EAS, PCS, or event dates verified against source order.
- No date conflicts or overlapping periods.
Content
- Every regulatory citation is the current version.
- Quantified impact figures match source documentation.
- No passive voice in citations, FITREPs, or correspondence.
- Word count matches requirement for the document type.
Format
- No [PLACEHOLDER] brackets remain in the output.
- Required enclosures listed and physically attached.
- Signature block correct (name, rank, billet, date).
- Output format matches required document type.
Security and Classification
- No PII hardcoded in the output (SSN, DOB, account numbers).
- CUI outputs banner-marked before printing or transmitting.
- Classification marking matches content sensitivity.
- Document routes only to personnel with need-to-know.
How to Use the Field Manual
Step 1. Choose Your Category
Match your task to one of the 10 categories.
Step 2. Select the Prompt
Read the prompt body before filling variables. Confirm the output is what you need. Check the classification tag.
Step 3. Fill Your Placeholders
Replace every [PLACEHOLDER] with your real value before sending. Never leave a bracket unfilled. Be specific.
Step 4. Attach a Persona Card (Optional)
For leadership and operational planning prompts, add a role. Example. "Act as a senior USMC staff officer with 20 years of administrative experience."
Step 5. Verify Your System
UNCLAS prompts work on any approved system. FOUO prompts require access-controlled systems. CUI prompts require a DoD-authorized system with a current ATO.
Step 6. Review and Score the Output
Check against five criteria. Clarity, constraint density, output specificity, variable coverage, regulatory accuracy. Score below 7/10 means iterate the prompt, not the regenerate button.
Step 7. Verify Facts Against Source Systems
Check every name, rank, date, and regulatory citation against MOL, MCTFS, or the governing MCO. AI output is a draft. Source systems are ground truth.
Step 8. Route Through Chain of Command
Replace all [PLACEHOLDER] variables with actual data in a secure document. Route through proper review and signature authority. You remain fully accountable.
Time Saved Benchmarks
The manual reports average time savings of 75 to 82 percent per document across the library. For an S-1 shop producing 20 admin actions per week, the system recovers 8 to 12 hours of manpower per week, equivalent to one admin billet of productive time returned to the unit. Time recovered does NOT eliminate verification. It front-loads drafting and frees SNCOs for review.
Audience
The Field Manual is built for three audiences.
Administrators
- 0111 Admin Clerks.
- S-1, S-3, S-4 Shop Chiefs.
- Unit Admin NCOs.
Leaders
- Commanding Officers.
- Executive Officers.
- SNCOs.
- Staff Officers.
Operators
- Operations NCOs.
- Training Officers.
- Logistics Coordinators.
Mission Packs (Multi-Prompt Bundles)
The manual organizes common multi-step tasks into Mission Packs that run prompts in sequence.
- EAS Separation Pack. Cat 02 + 09 + 08.
- NJP Pack. Cat 05 in order.
- Awards Pack. Cat 04.
- Change of Command. Cat 02 + 06 + 10.
- Pre-Deployment Pack. Cat 03 + 06 + 07.
- Disciplinary Action. Cat 05 + 02.
Cross References
- For overall AI framework, see AI Overview and Authorities
- For approved AI platform (GenAI.mil), see GenAI.mil Enterprise Platform
- For other authorized tools, see CamoGPT and Other Approved AI Tools
- For ethical use principles, see AI Ethical Use
- For mandatory training, see Basic AI Course (Mandatory)
- For day-to-day practical usage, see Practical AI Use for Marines
- For additional training options, see AI and Data Management Training
- For governance and approval, see AI Governance and Approval
Wiki Parent Groups Aligned to Field Manual Categories
- Cat 02 Personnel Administration aligns with Life Events, Reenlistments and Extensions, Separations and Retirement.
- Cat 03 Training Administration aligns with Education.
- Cat 04 Awards and Evaluations aligns with Awards and Performance Evaluation.
- Cat 05 Legal and Disciplinary aligns with Legal Services.
- Cat 08 Financial Administration aligns with Pay and Entitlements, DTS, GTCC.
- Cat 10 Leadership and Decision Making aligns with Marine and Family Programs.
Quick Reference
- Version. 3.0 (2026).
- Prompts. 100 across 10 categories.
- Pages. 53.
- Classification ceiling. CUI.
- Authority. NAVMC 5239.1, MARADMIN 018/26, EO 13556, 32 CFR Part 2002.
- Platform. GenAI.mil per MARADMIN 018/26.
- License. Built by Marines for Marines. Open and improving.
- Source. github.com/semperadmin.
Records To Keep
- Downloaded Field Manual PDF (current version)
- Awareness of the Semper Admin Standard five pillars
- Practiced verification habit per Trust but Verify checklist
- Documentation of any official products produced using Field Manual prompts
If you have a question about a specific prompt in the Field Manual, refer directly to the manual section for that category. To contribute corrections or improvements, visit github.com/semperadmin. To verify ATO status of your LLM system before processing CUI, contact your unit S-6 or DC I SDO.
How other roles handle this
- Admin view - the processing side
- Leader view - your NCO and SNCO oversight
- Commander view - command authority and decisions
References
- Semper Admin AI Field Manual v3.0 (2026) - github.com/semperadmin
- MARADMIN 018/26 (Enterprise GenAI Availability and Governance)
- MARADMIN 214/26 (Mandatory Basic AI Course)
- MARADMIN 635/24 (AI Ethical Use by COMMSTRAT)
- NAVMC 5239.1
- EO 13556 (Controlled Unclassified Information)
- 32 CFR Part 2002 (CUI Regulation)
- MCO 5231.4 (Marine Corps Data and Artificial Intelligence)
- NAVMC 3000.1 (USMC AI Implementation Plan, FY 2025-2030)
Related Pages