jonnifyagile agents
jonnify / Agile Agent Workflow / A5 · The agent brief / What the chapter covers

A5 · orientation dive 2

What the chapter covers

An .llms.md brief has five parts, in order: references, requirements, execution topology, agent stories, and the one implementation prompt. The eight modules of A5 teach each part, the practice of running an agent, the thesis, and a workshop that ships a first increment.

Each of the five parts has a role, a module that teaches it, and a real Portal artifact it lands on. The brief is read top to bottom by an agent: references first — links, so the agent reads sources before prose — then the numbered testable requirements, the runtime topology and task DAG, the agent stories with their acceptance gates, and finally the single prompt that runs the whole build in task order. Select a part on the right to read its role and where it lands.

The five parts of an .llms.md · pick one

one .llms.md · five parts, references first PART 1 · A5.2 References — every source the agent reads, links first PART 2 · A5.2 Requirements — numbered, testable, traced to a story PART 3 · A5.3 Execution topology — the runtime tree, the task DAG, the files PART 4 · A5.4 Agent stories — a Directive and an Acceptance gate, each PART 5 · A5.5 The implementation prompt — one run, in task order, to the gates
Part 1 of 5 · References — every source the agent reads, links first. Taught by A5.2 References and requirements. Lands on f6.1.llms.md — the Phoenix docs, the Portal facade contract, and the design system, links first.

The five parts

An .llms.md brief is one document with five named parts, read top to bottom. References — every source the agent reads, links first, so it reads sources before prose. Requirements — numbered, testable statements, each traced to a user story. Execution topology — the runtime process tree, the build-order task DAG, and the exact files touched. Agent stories — the executable counterpart of a user story, each a Directive the agent runs and an Acceptance gate that closes it. The implementation prompt — the single prompt an agent runs to build the increment in task order, ending on the gates.

Read in that order, the brief assembles a system rather than a pile of snippets: an agent reads the sources, learns the requirements, sees the shape and the order, takes each story to its gate, and runs the one prompt that composes the whole build.

1 · references
Every source the agent reads, links first. Taught by A5.2.
2 · requirements
Numbered, testable, each traced to a story and an invariant. Taught by A5.2.
3 · topology
The runtime tree, the build-order task DAG, the touched files. Taught by A5.3.
4 · agent stories
A Directive and an Acceptance gate, each closing one user story. Taught by A5.4.
5 · the prompt
One prompt, run in task order, self-checking against the gates. Taught by A5.5.

A5.1 precedes all five with the convention itself — why a brief for an agent is written links-first, prose-second — and A5.6–A5.8 cover the practice, the thesis, and a workshop that ships a first increment.

The eight modules

Eight teaching modules carry the chapter: the convention (A5.1), the five parts (A5.2–A5.5), the practice of running an agent (A5.6), the thesis (A5.7), and a workshop (A5.8). Select a module below and the readout reports its title, its one-line role, and the group it belongs to — the same orientation the landing's module grid gives, walked one module at a time.

The eight A5 modules · pick one

A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 A5.1 · Writing for an agent: the llms.txt convention group: the convention
A5.1 · Writing for an agent: the llms.txt convention — why a brief for an agent differs from documentation for a person: links first, prose second, every reference exact. Group: the convention. One of eight modules A5 teaches, shipped after the landing.

A5 is five parts taught across eight modules: the convention, each part, the practice, the thesis, and a workshop that ships a first increment.

On the Portal

The five parts are not an abstraction; they are the literal sections of the Portal's real briefs. The web bootstrap's f6.1.llms.md opens with References (the Phoenix docs, the Portal facade contract, the design system), then Requirements (F6.1-R1–R8), then an Execution topology (the runtime tree plus the T1–T7 task DAG and the touched files), then Agent stories (F6.1-AS1–AS4), and closes on the implementation prompt. Each A5 module teaches one of these against that real artifact.

# f6.1.llms.md — the five parts, in order (verbatim section headings)
  ## References              # links first: Phoenix docs · the Portal facade · f0.md
  ## Requirements            # F6.1-R1 … F6.1-R8, each traced to a user story
  ## Execution topology       # the runtime tree · the task DAG T1 → T7 · the touched files
  ## Agent stories            # F6.1-AS1 … AS4, each a Directive + an Acceptance gate
  ## Comprehensive implementation prompt   # the one prompt, run in task order, to the gates

The principle

A brief is one document with five named parts; reading it top to bottom assembles a system, not a pile of snippets — sources, then requirements, then shape and order, then stories, then the run.

On the Portal

f6.1.llms.md carries exactly those five parts — References, Requirements (R1–R8), Execution topology (T1–T7), Agent stories (AS1–AS4), and the implementation prompt.

The five parts are the literal sections of a real Portal brief; the eight modules teach them one at a time.

Recap — five parts, eight modules

A5 covers the five parts of an .llms.md — references, requirements, execution topology, agent stories, the implementation prompt — taught across eight modules: the convention, each part, the practice of running an agent, the thesis, and a workshop. Every part is a literal section of the Portal's real f6.1.llms.md. The next dive walks how you learn and build it here: write the brief part by part, run the agent, and review against the spec.

Next in A5: How you learn and build it — the method of writing the brief, running the agent, and reviewing the output against the spec's Definition of Done.

References

Primary sources for the machine-brief convention, structuring an agent's task, and the user-story form the agent story extends — and where the chapter landing and the spec it wraps sit in the course.

Sources

Related in this course