---
title: Shared-Context Pack
description: Structure of the shared-knowledge pack built by architect:implement-backlog
seo:
  image: /ogp.png
---
`architect:implement-backlog` builds this shared-knowledge pack first, before starting any implementation. Throughout implementation, every agent references this pack.
The following files are placed under `reports/backlog/shared-context/`.

| File | Contents | Generated from |
|---|---|---|
| `architecture-guardrails.md` | Package boundaries, transaction boundaries, prohibited technology choices, etc. | `architecture.md`, `tech-stack-fitness.md` |
| `coding-standards.md` | Naming conventions, exception-handling patterns, testing policy | Language/framework decisions |
| `ubiquitous-language.md` | Domain vocabulary glossary (original-language ↔ English) | `ubiquitous-language.md` (present in both product and architect artifacts) |
| `data-contracts.md` | Table definitions, aggregate boundaries, enum conversion rules | `data-model.md` |
| `nfr-budgets.md` | Quick-reference for NFR/SLA target values | `nfr.md`, `sla.md` |
| `decisions.md` | ADR-lite log of cross-cutting decisions (**append-only**) | Accumulated during implementation |
| `review-knowledge.md` | Distilled catalog of lessons learned during review (**append-only**, written by `review-issue`) | Repeated review rounds |

The diagram below shows which shared-context files are produced from each source document.

```mermaid
flowchart LR
  subgraph SRC["Sources"]
    A1[architecture.md]
    A2[tech-stack-fitness.md]
    A3["Language/framework decisions"]
    A4["ubiquitous-language.md<br/>(product/architect)"]
    A5[data-model.md]
    A6[nfr.md]
    A7[sla.md]
    A8["Accumulated during implementation"]
    A9["review-issue review rounds"]
  end
  subgraph SC["shared-context/"]
    B1[architecture-guardrails.md]
    B2[coding-standards.md]
    B3[ubiquitous-language.md]
    B4[data-contracts.md]
    B5[nfr-budgets.md]
    B6["decisions.md(append-only)"]
    B7["review-knowledge.md(append-only)"]
  end
  A1 --> B1
  A2 --> B1
  A3 --> B2
  A4 --> B3
  A5 --> B4
  A6 --> B5
  A7 --> B5
  A8 --> B6
  A9 --> B7
```

:::note[How append-only logs work]
`decisions.md` and `review-knowledge.md` are handled differently from the other files. These two are **append-only logs** — entries are only ever added; existing content is never deleted or overwritten. Unlike the other files, they are never regenerated from scratch.
`review-knowledge.md` in particular serves as a **mechanism to prevent the same problem from being implemented twice**. Lessons learned during review are accumulated without erasure.
For examples of rules that have been captured in practice, see the [review-issue page](/skills/backlog/review).
:::