一键在 Manus 中运行任何 Skill

factory-root-planner

Coordinate a Factory run from raw brief through Stage I2. Use when Codex is asked to initialize a Factory run, run Stage A, create or repair run-root evidence, enforce Factory read order, choose PLANNING_ONLY versus EXECUTION_ENABLED posture, coordinate the full stage flow, or prepare a pack for human review without executing implementation.

在 Manus 中运行

概览

安装命令

npx skills add https://github.com/EduardoRemedios/factory-starter-kit --skill factory-root-planner

复制此命令并粘贴到 Claude Code 中以安装该技能

来源

EduardoRemedios/factory-starter-kit

星标3

分支1

更新时间2026年4月26日 06:41

文件资源管理器

2 个文件

SKILL.md

readonly

name	factory-root-planner
description	Coordinate a Factory run from raw brief through Stage I2. Use when Codex is asked to initialize a Factory run, run Stage A, create or repair run-root evidence, enforce Factory read order, choose PLANNING_ONLY versus EXECUTION_ENABLED posture, coordinate the full stage flow, or prepare a pack for human review without executing implementation.

Factory Root Planner

Workflow

Read AGENTS.md, docs/Factory/ORCHESTRATION.md, docs/Factory/Spec/STAGE_CONTRACTS.md, and docs/Factory/SCRATCHPAD.md ## Active Pitfalls (Mandatory).
Run bash scripts/knowledge_lint.sh before Stage A and persist output as docs/Factory/runs/<RUN_ID>/KNOWLEDGE_LINT.txt.
Create the run root and required files: raw_brief.md, EXECUTION_MODE.txt, CONTEXT_RECALL_REPORT.md, and later SPRINT_ID.txt.
Build recall evidence with ./scripts/factoryctl context-index and ./scripts/factoryctl context-report --profile stage-a.
Keep the run PLANNING_ONLY unless the raw brief explicitly authorizes EXECUTION_ENABLED.
Coordinate stages in order: A -> B -> C -> D -> E -> F -> G -> H -> I -> J -> I2.
After each stage handoff, run ./scripts/factoryctl stage-lint --run <RUN_ID> --stage <STAGE>.
After I2, run ./scripts/factoryctl pack-lint --run <RUN_ID> before human review.

Guardrails

Do not execute sprint implementation during Factory planning.
Do not expand scope silently; record proposed expansion as BLOCKING unless approved.
Treat recall artifacts as evidence aids, not authority.
If Coverage Verdict: WEAK, halt and repair recall before drafting intent.
If any validator fails, fix the artifact or re-run the affected stage before advancing.

Outputs

Return:

run id
execution mode
evidence paths created
current stage status
validator results
blockers requiring human decision

同仓库更多 Skills

同仓库

factory-execution-closeout

EduardoRemedios/factory-starter-kit

Close implementation work against an approved Factory pack. Use when Codex is asked to verify execution after human Go, compare a code/doc diff to the approved envelope and micro-sprints, prepare execution completion notes, check residual risk, or decide whether a completed implementation is ready for project-specific merge gates.

2026-04-263

factory-pack-consolidator

EduardoRemedios/factory-starter-kit

Consolidate a Factory pack during Stage J. Use when Codex is asked to create or repair PACK_MANIFEST.md, create or repair PACK_CHECKLIST.md, verify required pack artifacts are present and non-empty before Stage I2, instantiate the Purple Gate checklist, or prepare a completed pack for pack-lint.

2026-04-263

factory-purple-gate

EduardoRemedios/factory-starter-kit

Perform Factory Purple Gate adjudication against evidence. Use when Codex is asked to review Stage D or Stage I2, decide PASS, CONDITIONAL PASS, or FAIL, evaluate PACK_CHECKLIST.md, resolve Red/Blue disputes, assess whether scope expansion remains unapproved, or produce PACK_AUDIT_REPORT.md from completed Factory artifacts.

2026-04-263

来源

EduardoRemedios

EduardoRemedios/factory-starter-kit

打开 GitHub 仓库查看创作者相关仓库

安装命令

下载

在 Manus 中运行

适用职业SOC

采购经理管理类职业11-3061L4

name	factory-root-planner
description	Coordinate a Factory run from raw brief through Stage I2. Use when Codex is asked to initialize a Factory run, run Stage A, create or repair run-root evidence, enforce Factory read order, choose PLANNING_ONLY versus EXECUTION_ENABLED posture, coordinate the full stage flow, or prepare a pack for human review without executing implementation.

Factory Root Planner

Workflow

Read AGENTS.md, docs/Factory/ORCHESTRATION.md, docs/Factory/Spec/STAGE_CONTRACTS.md, and docs/Factory/SCRATCHPAD.md ## Active Pitfalls (Mandatory).
Run bash scripts/knowledge_lint.sh before Stage A and persist output as docs/Factory/runs/<RUN_ID>/KNOWLEDGE_LINT.txt.
Create the run root and required files: raw_brief.md, EXECUTION_MODE.txt, CONTEXT_RECALL_REPORT.md, and later SPRINT_ID.txt.
Build recall evidence with ./scripts/factoryctl context-index and ./scripts/factoryctl context-report --profile stage-a.
Keep the run PLANNING_ONLY unless the raw brief explicitly authorizes EXECUTION_ENABLED.
Coordinate stages in order: A -> B -> C -> D -> E -> F -> G -> H -> I -> J -> I2.
After each stage handoff, run ./scripts/factoryctl stage-lint --run <RUN_ID> --stage <STAGE>.
After I2, run ./scripts/factoryctl pack-lint --run <RUN_ID> before human review.

Guardrails

Do not execute sprint implementation during Factory planning.
Do not expand scope silently; record proposed expansion as BLOCKING unless approved.
Treat recall artifacts as evidence aids, not authority.
If Coverage Verdict: WEAK, halt and repair recall before drafting intent.
If any validator fails, fix the artifact or re-run the affected stage before advancing.

Outputs

Return:

run id
execution mode
evidence paths created
current stage status
validator results
blockers requiring human decision