init-output
Create the architect output directories and work/pipeline-progress.json for a new project.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 13
- Forks
- 1
01 前言 Front Matter
- model
- haiku
- description
- Create the architect output directories and work/pipeline-progress.json for a new project.
- argument-hint
- [project_name]
03 SKILL.md 正文
Output Initialization
Shared files: a path written @rules/…, @skills/…, @templates/… or @docs/… is relative to the
plugin root, ${CLAUDE_PLUGIN_ROOT} — not to the project being worked on. Read it from there.
Usage: /architect:init-output [project_name]
Use --reset to reinitialize.
Expected Outcome
Create the directory structure and progress management files required for pipeline execution.
Existing State Is Never Discarded
This skill is additive. A project may already carry state when it runs — most often
because the product pipeline ran first and this is the product→architect handoff
(@docs/design.md §1.1–1.5), but equally on any re-run of the architect path. Read each file
before writing it and merge into what is there. Only --reset replaces anything, and only
after a backup.
What is at stake: work/context.md is the Open Questions store for the whole project — both
pipelines' questions, in one file (@rules/open-questions.md §6) — which
/architect:define-requirements reads in its Step 2 to re-ask the deferred / unasked entries
it needs and answer them in place; work/pipeline-progress.json is shared
by both pipelines and already holds the product phases and the output_language the user
chose during the product run.
Execution Steps
-
Create the following directories:
reports/before/{project}/reports/00_summary/reports/01_analysis/reports/02_evaluation/reports/03_design/reports/review/individual/generated/work/
-
Create or merge
work/pipeline-progress.json:- Absent — create it, registering every phase from
@skills/common/skill-dependencies.yamlas"pending", each with the fields defined in @skills/common/progress-registry.md — including"plugin": "architect", which is what makes an entry attributable once both pipelines share the file. - Present — keep the file and add only the phase entries it does not already have, as
"pending". Never reset an entry that exists, never remove a phase this manifest does not define (the product pipeline registers its own phases in the same file), and leavegates,errors,warningsand any other top-level key untouched. options— keep every value already set, in particularoutput_language, which the product run may have asked the user for. Fill in only what is missing:output_languagedefaults to"en",confirm_versionstotrue(whether codegen skills confirm resolved dependency versions with the user — see @rules/dependency-versions.md),eli5to"technical"(which reports open with the plain-language section of @rules/eli5-summary.md).- Colliding phase names. Four names are defined by both manifests —
map-domains,design-api,create-domain-story,report— and the registry keys phases by bare name, so an existing entry under one of them may describe the product phase rather than this one. Leave it exactly as it is. If it carries"plugin": "product"the question is settled and there is nothing to report; if it carries nopluginat all and is alreadycompleted, append one line towarnings[]naming it and stating that its status may belong to the product pipeline, so the architect phase is unverified. The status dashboard renderswarnings[], so the ambiguity is visible instead of reading as done. Never add"plugin": "architect"to an entry you did not create — that would claim the neighbour's work as this pipeline's.
- Absent — create it, registering every phase from
-
Create
work/context.mdonly if it is absent, seeded with an empty## Open Questionssection whose table has the @rules/open-questions.md §6 header, verbatim:## Open Questions | ID | Question | Status | Answer | Options offered | Owner | Impact | Asked at | |----|----------|--------|--------|-----------------|-------|--------|----------|If it exists, leave its content in place: phases append to it, and it is the Open Questions store for the whole project — both pipelines' questions live in this one file, so on the handoff path it already holds the entries
/architect:define-requirementsis about to read, re-ask and answer in place.reports/00_requirements/open-questions.mdis a view rendered from it, never a second store.
Options
--reset: Back up the existingwork/pipeline-progress.json(copy to*.bak), then re-register this manifest's phases as"pending"— phases belonging to the product pipeline are still preserved.--resetdoes not touchwork/context.mdorwork/traceability.json; both are shared with the product pipeline.
Completion Criteria
The directory structure and work/pipeline-progress.json exist, every architect phase has an
entry, and no pre-existing phase entry, option value, or work/context.md content was lost.
04 同仓库更多 Skills
verify-sql-migration
Prove a SQL migration on data: golden checks against an authorized source database and differential tests against ScalarDB. Never touches production.
verify-implementation
Check code against the design (API contract, transactions, security, requirements, the example chain) and report every divergence; optionally runs the eight-stage quality gate.
update-knowledge
Fetch or update an OKF knowledge bundle: ScalarDB/ScalarDL, or the vendored Kubernetes/Terraform bundle.
start
Interactive entry point for system architecture work: refactor a legacy application into services, or design a new system from requirements; picks the path from the project's state. To move a database to ScalarDB, use migrate-database.
select-scalardb-edition
Choose the ScalarDB edition (Community, Enterprise Standard or Premium, plus Enterprise Options), deployment mode and cluster topology through a short interview.
scaffold
Generate a runnable ScalarDB starter project for any of the six interface combinations: build, config, schema, service class, docker-compose and README.