git:20260909.e845f00 to git:20260925.3e2b90a

- ArchHarness is a Claude Code skill pack for enterprise architecture work.

- It turns Claude into a team of architecture specialists you summon on demand.

- ## Available skills

- | Skill | Command | Role |

- |-------|---------|------|

- | arch-workflow | `/arch-workflow` | Pipeline gatekeeper — enforces stage order and PASS/WARN/BLOCK gate; blocks skipping |

- | arch-requirements | `/arch-requirements` | Orchestrator — interview + multi-source intake, outputs REQ.md + req.yaml |

- | arch-req-from-diagram | `/arch-req-from-diagram` | Reader — draw.io / D2 / arch YAML / PNG (vision) → partial req.yaml |

- | arch-req-from-doc | `/arch-req-from-doc` | Reader — PDF / DOCX / MD / TXT via LLM → partial req.yaml |

- | arch-req-from-api | `/arch-req-from-api` | Reader — CMDB / ServiceNow / CSV → partial req.yaml |

- | arch-req-merge | `/arch-req-merge` | Merger — combine partials, conflict detection, gap report | Requirements specialist — structured interview, outputs REQ.md + req.yaml for arch-design |

- | arch-validate | `/arch-validate` | Paranoid security architect — validates a diagram image against all rules |

- | arch-design | `/arch-design` | Senior architect — designs from requirements, picks patterns, generates YAML |

- | arch-diagram | `/arch-diagram` | Diagram generator — converts Architecture YAML → draw.io file + PNG |

- | arch-enforce | `/arch-enforce` | CI enforcement gate — applies arch-gate-policy.yaml to validate JSON, emits PASS/WARN/BLOCK |

- | arch-security | `/arch-security` | Security auditor — focused exclusively on auth, credentials, network boundaries |

- | arch-review | `/arch-review` | Architecture committee reviewer — checks standard compliance, scores dimensions |

- | arch-report | `/arch-report` | Technical writer — generates executive summaries and Confluence-ready docs |

- | arch-optimize | `/arch-optimize` | Staff architect — identifies improvements, generates prioritized fix list |

- All company-specific values live in **`config.yaml`** at the project root.

- Edit it before first use — this is the only file you need to change.

-

- Key fields:

- - `company.name` — used in report headers and classification labels

- - `datacenters` — your DC names, locations, and network zones

- - `platforms` — API gateway, message bus, K8s platform names

-

- ```yaml

- # config.yaml (excerpt)

- company:

- name: "Acme Corp"

- datacenters:

- - id: "dc-primary"

- aliases: ["Primary DC"]

- location: { city: "Tokyo", country: "JP" }

- zones: ["DMZ", "App Zone", "DB Zone"]

- platforms:

- api_gateway: "Kong API Gateway"

- message_bus: "RabbitMQ"

- ```

-

- Inputs and outputs are managed per project under `projects/<id>/` — not via

- top-level `input/`/`output/` folders.

-

- ## Diagram generation tool

-

- `tools/arch-diagram-gen/` — Python tool that converts Architecture YAML to draw.io XML and PNG.

-

- ```bash

- # Install

- pip install -e ".[all]"

-

- # Generate (output goes to the active project output/diagrams, or ./ relative to cwd)

- python tools/arch-diagram-gen/arch_diagram_gen.py -i arch.yaml

-

- # Equivalent, and directory-independent once `archharness` is on PATH:

- python -m archharness diagram -i arch.yaml

- # Override output explicitly

- python tools/arch-diagram-gen/arch_diagram_gen.py -i arch.yaml -o out.drawio --png out.png

- # Target a specific workspace project

- python tools/arch-diagram-gen/arch_diagram_gen.py -i arch.yaml --project payments

- ```

- See `tools/arch-diagram-gen/README.md` and `arch-schema-reference.yaml` for full docs.

- The tool produces standard architecture shapes: hexagons for F5/FW, parallelograms for

- API gateways, cylinders for databases, dashed zones, status colors.

- All skills load company-specific values from `config.yaml` at runtime.

- The `standards/` directory holds the platform-agnostic rules and topology requirements:

- - `platform-rules.yaml` — AWS/Azure/private-cloud platform-specific rules

- 1. **Design** → `/arch-design` to generate architecture YAML from requirements

- 2. **Validate** → `/arch-validate` with diagram image to get scored JSON report

- 3. **Enforce** → `/arch-enforce` to apply the CI gate policy (or skip for human review)

- 4. **Deep-dive security** → `/arch-security` for auth/credential/network boundary audit

- 5. **Standards check** → `/arch-review` for committee-style compliance scoring

- 6. **Fix it** → `/arch-optimize` for prioritized improvement suggestions

- 7. **Document it** → `/arch-report` for executive summary or Confluence page

-

- ## If skills aren't loading

-

- Check that `.claude/skills/` is on the project path. Skills follow the

- Claude Code Agent Skills open standard — each directory under `.claude/skills/`

- with a `SKILL.md` is automatically registered as a slash command.

-

- ## Multi-project workspace

-

- One ArchHarness checkout supports many isolated architecture projects.

- Initialize once, then create one project per system/integration:

-

- ```bash

- python -m archharness init-workspace .

- python -m archharness init-project <id> [--name "..."] [--default]

- python -m archharness list-projects

- python -m archharness doctor [--project <id>]

- ```

-

- Each project owns `projects/<id>/{input,working,output}` plus a `project.yaml`

- (`id`, `name`, `platform`, `data_classification`). When you open a session inside

- `projects/<id>/`, skills and CLI tools (`arch_diagram_gen.py`,

- `tools/arch-req-readers/req_reader.py`) auto-detect the active project and write

- into that project's `output/`. `--project <id>` selects explicitly from anywhere

- in the workspace. Project data dirs are git-ignored; only `project.yaml` and the

- project `README.md` are tracked.

- Stages must run in the order defined by `standards/workflow.yaml`:

- findings and re-running validate → enforce.

- - Ask `/arch-workflow status` or `/arch-workflow can <stage>` when in doubt.

- ## OpenCode usage

- This project also supports OpenCode. Use `@agent-name` instead of `/skill-name`:

- @arch-workflow → check stage gate / pipeline status (gatekeeper)

- @arch-validate → validate a diagram image

- @arch-design → design from requirements

- @arch-enforce → CI gate decision (PASS/WARN/BLOCK)

- @arch-security → security deep-dive

- @arch-review → committee gate decision

- @arch-optimize → prioritized fix backlog

- @arch-report → Confluence page / exec summary

- Agent definitions live in `.opencode/agents/`. They point to the same

- `SKILL.md` files in `.claude/skills/` that Claude Code uses.

- The `AGENTS.md` in this directory is the OpenCode-native equivalent of this file.