How to Meet Government Terms of Reference with AI Agents: A 2026 Guide
About this translation
Original publication dates and historical context are preserved. Images, embedded media, source documents and code examples remain in their original language.

If you provide services to the government, you already know that winning the contract is only part of the job. Payment comes when the requesting department formally accepts the work, and to do so it checks each deliverable against the terms of reference (TDR, the Spanish acronym).
In this guide, I show you how to organize a public-sector contract with AI agents in Claude so that every task, meeting and deliverable is linked to the TDR clause it addresses. I designed it for construction, technology, legal, communications, marketing and consulting businesses and consultants working with public entities in Peru.
In brief
- The TDR is the reference the entity uses to review your work, so it makes sense to organize the entire contract around it.
- A standalone chat loses context. A living contract file with five agents maintains traceability from signing through formal acceptance.
- Projects, Skills, Cowork, Claude Code, MCP connectors, Granola and Obsidian perform different roles within the same workflow.
- Law 32069, Law 31814 and Law 29733 set clear boundaries for the information you upload, your responsibility and human oversight.
Why So Many Government Deliverables Receive Review Comments
A deliverable rarely receives review comments because the team did poor work. More often, the team did good work on something the TDR required in a different form, or failed to address a clause buried in an annex.
These situations recur in almost every industry:
- Requirements are scattered across the TDR, its annexes, the consolidated bidding documents, the bid and the contract.
- Strict compliance obligations become mixed up with enhancements the supplier offered.
- Coordination meetings produce new requests that are not in the TDR and that nobody records in writing.
- Information is dispersed across emails, meeting minutes, WhatsApp messages and files with different names.
- Team members change, and newcomers do not know about earlier agreements.
The result is familiar: review comments, rework, delayed acceptance and, in the worst case, penalties for late performance.
There is also a legal framework that keeps changing. Law No. 32069, the General Public Procurement Law, has been in force since April 22, 2025, with its regulations approved by Supreme Decree No. 009-2025-EF. The regulations were amended in January 2026 by Supreme Decree No. 001-2026-EF, and in July 2026 Law No. 32732 introduced further changes to the law. It is therefore worth checking which version of the bidding documents and contract applies to each procurement process.
Why a Standalone Chat Is Not Enough for a Public-Sector Contract
Many teams already use AI, but in disconnected pieces: one prompt to summarize the TDR, another to draft a report, another to reply to an email. Every conversation starts from scratch.
This creates three problems. The first is context: the tool does not remember what was agreed last week or what changed in the contract amendment. The second is traceability: the text may be well written yet fail to identify the clause it addresses. The third is information handling: contract documents end up being copied into tools without any controls.
Agents address this when they work from the same information base and each has a narrowly defined task. For a broader overview of how they work, see this guide to AI agents for B2B businesses.
Build a Living File for Each Contract
The living contract file is a single folder containing everything that defines the contract. It can be a Project in Claude or an Obsidian vault shared with your team. What matters is that agents always read from it rather than from disconnected copies.
- Core documents: TDR, annexes, consolidated bidding documents, submitted bid, contract and amendments.
- Obligations matrix: the table linking each obligation to its clause, deliverable, deadline and verification method.
- Activity log: meeting minutes, agreements and relevant emails, ordered by date.
- Deliverables: each submitted version, the review comments received and the response to each one.
One simple rule helps a great deal: name files using the deliverable number and the TDR clause they address. That way, any team member or agent can immediately find the supporting evidence.
Five Agents That Support the Contract from Signing to Acceptance
Each agent has a task and a specific output. Between agents, there are points where the team reviews the work and makes decisions.
- 1
TDR readerTurns the TDR, contract and amendments into the obligations matrix. Everything else builds on it.
- 2
PlannerUses the matrix to create the deliverables schedule, with Google Calendar alerts through the connector.
- 3
Meeting reporterUses the Granola transcript to draft meeting minutes and check whether each request is within scope.
- 4
Deliverable writerWrites each deliverable in the format required by the TDR, supported by a skill and the evidence in the contract file.
- 5
Acceptance reviewerReads the draft as the requesting department would and anticipates review comments before submission.
The team intervenes at two points: it validates the matrix before planning and approves each deliverable after review.
The TDR Reader Builds the Obligations Matrix
The matrix is the contract's most important document. Reviewing it with the project manager before moving forward takes little time and can save weeks of rework.
TDR reader prompt
Actúa como gestor de contratos de un proveedor del Estado.
Lee el TDR completo, incluidos los anexos, y el contrato firmado.
Arma una matriz de obligaciones con estas columnas:
N.º | Obligación | Numeral del TDR | Entregable asociado | Plazo | Penalidad aplicable | Cómo la verificará el área usuaria
Reglas:
- Incluye solo obligaciones que puedas citar con su numeral.
- Si el contrato o una adenda cambia algo del TDR, márcalo y cita ambos documentos.
- Distingue lo que exige el TDR de lo que ofrecimos como mejora en la oferta.
- Al final, lista las dudas que conviene aclarar con la entidad por escrito.
The Planner Adds Deadlines to the Calendar
With the Google Calendar connector, the agent records each delivery date with advance alerts. Count deadlines as the contract defines them, in calendar days or working days, and have a person validate the initial schedule. After that, deadlines no longer depend on someone on the team remembering them.
The Meeting Reporter Separates Requests from What the TDR Covers
Meetings with the entity involve people with different roles: the requesting department, the procurement unit and, sometimes, a supervising party. Each may ask for different things. Granola transcribes the meeting, and the agent checks each request against the matrix.
Meeting reporter prompt
Actúa como gestor de contrato de un proveedor del Estado.
Insumos: la transcripción de la reunión con la entidad y la matriz de obligaciones del TDR.
Entrega:
1. Acta breve: fecha, asistentes por cargo, acuerdos y responsables.
2. Tabla de pedidos con estas columnas:
Pedido | Quién lo pidió (cargo) | Numeral del TDR que lo sustenta | Dentro o fuera del alcance | Impacto en plazo o costo
3. Lista de pedidos fuera del alcance que conviene formalizar por escrito antes de ejecutarlos.
Reglas:
- Si no encuentras el numeral, marca "sin sustento en el TDR". No lo deduzcas.
- No incluyas datos personales de los asistentes; usa solo cargos.
When a request falls outside the scope, it is advisable to formalize it in writing before carrying it out. This prevents the team from taking on work that nobody later acknowledges.
The Writer Uses a Skill for Each Document Type
A skill is a set of instructions Claude applies whenever you perform a particular type of task. For a public-sector contract, it helps to have one for each recurring document: monthly report, meeting minutes, letter to the entity and response to review comments. Here is a simple example:
Example skill for monthly reports (SKILL.md)
---
name: informe-mensual-contrato
description: Redacta el informe mensual del contrato siguiendo el formato del TDR. Úsalo cuando el usuario pida el informe mensual o un avance de entregable.
---
# Informe mensual del contrato
1. Lee la matriz de obligaciones y la bitácora del mes.
2. Sigue el orden de secciones que exige el numeral del TDR sobre informes.
3. Cada avance que reportes debe citar su evidencia: acta, correo o archivo.
4. Cierra con una tabla de pendientes, responsables y fechas.
5. Si falta evidencia para un punto, márcalo como "pendiente de sustento".
The Reviewer Reads the Deliverable as the Entity Will
Before submission, an agent other than the writer checks the draft against the matrix. Using a different agent helps identify requirements the writer assumed were fulfilled without verifying them.
Acceptance reviewer prompt
Actúa como el especialista del área usuaria que debe dar la conformidad.
Insumos: la matriz de obligaciones y el borrador del entregable N.º [número].
Revisa el borrador numeral por numeral y entrega una tabla:
Numeral | Qué exige el TDR | Dónde lo cumple el borrador (sección y página) | Cumple, parcial o no cumple | Qué falta
Después, redacta las tres observaciones más probables que haría la entidad y cómo resolverlas.
No corrijas el texto todavía. Solo evalúa.
Which Tool to Use for Each Part of the Work
| Tool | Its role in the contract |
|---|---|
| Projects | One project per contract containing the TDR, bidding documents, bid, contract and team instructions. It is the context everyone shares. |
| Skills | Recurring formats: monthly report, meeting minutes, letter to the entity, compliance matrix and response to review comments. |
| Plugins | An industry-specific package of skills, commands and connectors, ready to install for each new contract. |
| Cowork | Agents that work with contract folders and files from the desktop, without writing code. |
| Claude Code | Automation for technical users: create each contract's structure, review multiple deliverables at once and compare versions against the TDR. |
| MCP connectors | Access to Gmail, Google Drive and Google Calendar to read emails from the entity, save deliverables and record deadlines. |
| Granola | Transcription of meetings with the entity, which then feeds into meeting minutes and scope review. |
| Obsidian | A contract activity log in Markdown files, linked by clause and easy for agents to read. |
You do not need to code to get started. Projects, Skills and Cowork cover most of the workflow. Claude Code makes sense when the team wants to automate repetitive tasks, such as preparing the folder for each new contract or reviewing several deliverables in one run. If you have never worked with repositories, this guide to GitHub for non-programmers helps you take that step.
How the Workflow Adapts to Each Industry
| Industry | What the TDR typically requires | Where agents help most | What to watch out for |
|---|---|---|---|
| Construction (consulting and works supervision) | Periodic reports, technical reports, quantity takeoffs and drawings | Consolidating reports and checking their consistency with the technical project file | The signature and responsibility of the licensed professional |
| Technology | Functional requirements, service levels and acceptance records | Tracing each requirement to its test and evidence | Keeping the entity's credentials and data out of the tool |
| Legal services | Reports, responses to inquiries and deadline tracking | Deadline tracking and first drafts with cited sources | Professional confidentiality and verification of every legal citation |
| Communications and marketing | Plans, creative materials, coverage and metrics reports | Metrics reporting and adapting materials for each channel | Copyright, image rights and transparency about AI-generated content |
| Consulting | Assessments, methodology and reports for each deliverable | Maintaining the methodology offered in every report | Ensuring the deliverable does not depart from what was offered in the proposal |
What the Law Says and How It Affects Your Use of AI
Using AI does not change your obligations as a supplier. It does require you to take care of three things: the information you upload, responsibility for what you deliver and human oversight of every output.
| Legal instrument | What it establishes | What it means when working with agents |
|---|---|---|
| Law No. 32069 and its regulations (Supreme Decree No. 009-2025-EF, amended by Supreme Decree No. 001-2026-EF) | The public procurement framework in force since April 22, 2025. OECE took over the functions of the former OSCE. | The TDR, bidding documents and contract remain the reference. The information you submit must be truthful, whether prepared by a person or a tool. |
| Law No. 32732 | Introduced amendments to Law 32069 in July 2026. | Check which version of the law applies to your procurement process before preparing a bid or deliverable. |
| Law No. 31814 and its regulations (Supreme Decree No. 115-2025-PCM) | In force since January 22, 2026, with phased obligations by sector. It includes principles of privacy, respect for copyright, human oversight, transparency and accountability. | Keep a record of what the tool did and what the team reviewed. If your business develops or implements AI systems in healthcare, education, justice, security, or economics and finance, review the compliance deadlines, which started falling due in September 2026. |
| Law No. 29733 and its regulations (Supreme Decree No. 016-2024-JUS) | Personal data protection. | Anonymize names, identification numbers and citizens' data before uploading files. Review the tool's privacy settings and use enterprise plans when handling sensitive information. |
| Contract clauses | Confidentiality, ownership of deliverables and, in some cases, permitted tools. | Read them before uploading the first document. If you have questions, ask the entity in writing. |
This guide provides a practical framework and does not replace legal advice on your particular situation.
What Should Stay in Your Team's Hands
- Price, scope and any decision that changes what was offered.
- Signatures and opinions that require a licensed professional.
- Sworn declarations and the experience being substantiated.
- Sending letters and formal communications to the entity.
- The final decision on whether a deliverable is ready to submit.
Agents prepare, organize and review, but the supplier remains accountable to the entity.
How to Start with Your Next Contract
- 1.Create a Project for the contract and upload the TDR, bidding documents, bid and contract.
- 2.Review the confidentiality clauses and anonymize information as necessary before uploading any file.
- 3.Generate the obligations matrix and validate it with the project manager.
- 4.Add deadlines to the calendar with advance alerts.
- 5.Transcribe meetings with the entity and have the meeting reporter process each set of minutes.
- 6.Create a skill for the most frequently recurring deliverable.
- 7.Ask the acceptance reviewer to read each deliverable before you submit it.
What Changes When the Contract Revolves Around the TDR
When every contract document can be traced to a clause, conversations with the entity change. Review comments are answered with evidence, new requests are discussed in terms of scope, and the team stops relying on one person's memory. AI speeds up writing, but the most noticeable day-to-day difference is better organization.
Does Your Team Work with the Government?
If you want to implement this workflow with your own TDR documents, adapt the skills to your industry or train your team to work with agents, let's discuss your situation.
Frequently Asked Questions
Can I use AI to prepare deliverables for a government contract?
Generally, yes. The entity assesses whether the deliverable complies with the TDR and whether the information is truthful. Before starting, check whether the contract includes confidentiality clauses or restrictions on the use of tools.
Do I have to disclose that I used AI?
It depends on the contract and the entity. If required, include it in the methodology. If not required, it is still advisable to keep an internal record of what the tool did and what the team reviewed, in line with the transparency and human oversight principles in the regulations for Law 31814.
What information should I avoid uploading to an AI tool?
Personal data of citizens or public officials, access credentials and any document the contract classifies as confidential. If you need to work with that material, anonymize it first.
Do I need to know how to code to use agents in Claude?
No. Projects, Skills and Cowork cover most of the workflow without code. Claude Code adds value when you want to automate repetitive tasks.
Does this also work for small-value contracts?
Yes. A small-value contract also has requirements and deliverables that the entity verifies. In that case, the workflow can be reduced to the obligations matrix and the acceptance reviewer.
Sources
- OECE. Law No. 32069, the General Public Procurement Law, and Its Regulations. Gob.pe platform, version updated for 2026.
- El Peruano Official Gazette. Supreme Decree No. 001-2026-EF, Amending the Regulations for Law No. 32069. January 8, 2026.
- PRCP Abogados. Lexmail on Law No. 32732. July 2026.
- Cuatrecasas. What Do the Regulations for the Artificial Intelligence Law Establish? September 2025.
- CMS. The First Compliance Deadline for Law No. 31814 Falls on September 10. September 2026.
- Law No. 29733, the Personal Data Protection Law, and its regulations approved by Supreme Decree No. 016-2024-JUS.