MODULE 3 | TAILOR

/INIT AND AGENTS.MD

i. Teaching Bob your codebase

Across the first two modules (Assist and Delegate), Bob was put to work with minimal-to-no background or context about your codebase. Bob proved itself to be capable, but operating and working only from what it could read and infer based in the moment. Module 3 (Tailor) will change that approach markedly. Here you will teach Bob about your project and your standards, so that Bob begins behaving like a teammate who already knows how your team works.

There are three ways to tailor Bob which build on one another:

Provide Bob with durable knowledge about your codebase (so it stops rediscovering your project from scratch with every invocation)
Encode the standards Bob must always follow (so quality and conventions are enforced without needing to be re-typed)
Shape Bob's behavior to fit each role on your team (so a product manager and a senior engineer each get the specialized version of Bob that they need).

Notice where control lives at this stage in the AI Maturity Curve. Whereas in Modules 1 and 2 you approved Bob's work as it happened, here in Module 3 you write the rules down— commit them to the repository —and have them applied automatically, every time, for everyone. Viewed another way: Bob, at this stage in your jouney, has progressed beyond delegated work towards tailored work. Bob's behaviors, capabilities, and understanding are now being attuned specifically to the workflow and personas engaging with it.

ii. Why this matters

Large language models (LLM) are stateless and therefore every new conversation, without intervening strategies or techniques, will begin with no memory of the last interaction. Left to its own devices, Bob would need to rediscover your codebase from scratch at the start of each interaction — an inefficient approach for a small project and an impractically cumbersome approach for a large one.

The /init command solves this challenge. It scans your project and generates a structured summary that Bob loads as persistent context for every new conversation. The /init command thus becomes the primary mechanism for giving Bob durable knowledge of your project's structure, conventions, and code. All without needing to re-read every file with each new invocation.

RUNNING /init

Initializing a project's context takes seconds and requires no additional configuration. To do so using IBM Bob:

Open the agentic sidebar in Bob IDE
Enter the slash command: /init
Bob reads the relevant files in the project
Bob generates the agents.md context file

You'll have plenty of opportunities to practice this yourself when the hands-on portion of Module 3 kicks off. Beforehand, however, it's important to understand more about the complementary agents.md file.

THE AGENTS.MD FILE

The primary output of the /init command is agents.md: a plain-text Markdown file placed in the project root directory.

Think of agents.md as onboarding documentation written for specifically for Bob. It provides a concise, actionable summary of everything Bob needs to understand about the project in order to work effectively with it. The documentation typically covers: the project's overview and purpose; its directory structure and key file locations; the technology stack and dependencies; architectural patterns and conventions; and development workflows.

From the point it is first introduced and defined, agents.md is automatically prepended to every new conversation a user has with Bob, ensuring the LLM has immediate project context without requiring any manual file read efforts.

MODE-SPECIFIC CONTEXT FILES

Alongside the main agents.md Markdown documentation, the /init command also generates context files tuned to each one of Bob's built-in Modes (which we examined in Module 2):

FILE	MODE
`.bob/rules-agent/AGENTS.md`	Agent mode (formerly Code and Advanced modes)
`.bob/rules-ask/AGENTS.md`	Ask mode
`.bob/rules-plan/AGENTS.md`	Plan mode

Each mode-specific file extends the base context with what matters most for the successful operations of a given Mode. Context for Plan mode might emphasize architectural conventions, while Agent mode context would benefit from highlighting file structure and naming patterns.

iii. Hands-on: generating your project context

The larger your codebase, the more valuable a well-maintained AGENTS.md asset becomes. On a small project, Bob could conceivably read every file each conversation; even so, having a pre-defined AGENTS.md is still more efficient than the read-everything alternative. On a large project, an AGENTS.md manifest becomes the backbone of the entire workflow. It provides detailed guidance for how Bob should navigate the codebase, without exhausting the context window on file reads.

TIPS FOR /INIT AND AGENTS.MD

Execute the /init command after significant changes to ensure Bob picks up on any new modules or services, restructured directory trees, new framework, and so on. If there are instances where you (the administrator) want to capture context without performing an automated scan— or believe that an automated scan might miss some key elements, such as business rules or deployment conventions —then you may also edit the agents.md asset by-hand.

Let's give Bob durable knowledge of the Galaxium Travels codebase by running the /init command and generating its AGENTS.md files.

First, ensure that Bob has enabled Agent mode within the Agentic Sidebar.

Within the Agentic Sidebar, type the following and hit Enter to execute:
```
/init
```
You may need to click Approve once (once or several times...) to permission access to the agents.md file before this summary appears.

QUIZ QUESTION - IBM BOB LEVEL 4

If you are planning to pursue IBM Bob Level 4 accreditation, what is rendered on-screen or returned within a console as a result of this step are likely referenced as part of a multiple-choice question. It is strongly recommend that you record to a notebook (for reference later) the names and directory locaton of the three mode-specific context files inside .bob that are created as a result of running the /init command.

AGENTS OF GALAXIUM TRAVELS

After execution of the /init command, Bob's Agent mode will examine key front-end and back-end (service) files within the Project directory, ultimately writing a total of four files.

AGENTS.md — General guidance

Commands require running from specific subdirectories (pytest from booking_system_backend/, npm from booking_system_frontend/)
Single-test syntax with full pytest class path
The dual REST+MCP architecture on a single port

.bob/rules-agent/AGENTS.md — Coding gotchas

No seats_available column — test fixtures in the repo use an old column name that no longer exists in the model; discovered only by reading both models.py and the tests
MCP tools bypass FastAPI DI — they call SessionLocal() directly, so tests must patch two places
verbatimModuleSyntax enforces import type for type-only imports
Cancellation is POST, not DELETE

QUIZ QUESTION - IBM BOB LEVEL 4

If you are planning to pursue IBM Bob Level 4 accreditation, what is rendered on-screen or returned within a console as a result of this step are likely referenced as part of a multiple-choice question. It is strongly recommend that you record to a notebook (for reference later) details about the "gotchas" associated with a cancellation (it's a POST, not a DELETE).

.bob/rules-ask/AGENTS.md — Context for questions

Backend tests are partially outdated (use the old seat model)
Frontend has zero tests
booking_time has no timezone suffix despite being UTC

.bob/rules-plan/AGENTS.md — Architecture constraints

Price multipliers are frontend-only — backend stores only the economy base price
SQLite with no migration framework (no Alembic)
seed.py runs on every startup — not safe for multi-worker deployments
No update endpoint for bookings — cancel+rebook is the only flow

iv. Key takeaways

LLMs are stateless by default: every conversation starts fresh, with no memory of the last session or chat
The /init command scans your project(s) and generates agents.md documentation, a context file that will subsequently be included with every new Bob conversation
AGENTS.md functions like onboarding documentation: a structured summary that Bob utilizes to navigate your project(s)
Mode-specific AGENTS.md files give targeted context for IBM Bob's Agent, Plan, and Ask modes
For large-scale projects, AGENTS.md is essential: Bob cannot feasibly read every file at the start of each conversation without exhausting its (sizeable) context window
Re-run the /init command after major changes to capture the latest context for project(s)

In the chapter ahead, you will examine strategies for encoding your organization's coding standards using Bob Rules.