role-blueprint
Produce a machine-checkable Agent Role blueprint before writing Role source; use after research, candidate scoring, or source ingestion to define Role id, aliases, catalog level, memory, skills, references, tools, adapters, permissions, provenance, validation plan, write scope, and single-role/multi-role/topology decision.
- License 许可
- Apache-2.0
- 最近更新
- 2026-08-14
- Stars
- ★ 19
- Forks
- 2
01 前言 Front Matter
- name
- role-blueprint
- description
- Produce a machine-checkable Agent Role blueprint before writing Role source; use after research, candidate scoring, or source ingestion to define Role id, aliases, catalog level, memory, skills, references, tools, adapters, permissions, provenance, validation plan, write scope, and single-role/multi-role/topology decision.
03 SKILL.md 正文
Role Blueprint
Use this skill before creating or patching Role source. A blueprint is the
write gate: no roles/<id>/ scaffold should be created until the Role shape,
contents, boundaries, and validation plan are explicit.
Inputs
- Research brief and research evidence.
- Candidate scorecard and selected source decision.
- Source inventory from
role-source-ingestwhen external source is involved. - Intended Role id, aliases, catalog level, publication target, host adapters, and permission posture.
- User constraints and unresolved decisions.
Workflow
- Decide shape.
- Choose
single_role,multiple_roles, ortopology_recipe. - If the user requires a single Role despite multi-agent source, list the tradeoffs and unsupported surfaces.
- Choose
- Define identity.
- Role id, name, aliases, purpose, responsibilities, non-goals, interaction mode, action posture, catalog level, and version strategy.
- Map contents.
- Memory, skills, references, prompts, tools, plugin content, adapters, templates, schemas, and tests.
- Mark
vendored_intact,vendored_modified,synthesized,referenced_only, andexcludedmaterial. Oldercopiedrecords remain readable as legacy provenance.
- Define boundaries.
- Project Binding, runtime state, provider state, generated projection output, secrets, and hidden installer state must stay outside Role source.
- For vendored public/open-source skills, record upstream ref, license, required notices, whether the content was modified, and the paths carried into the Role package.
- Define validation.
- TOML parsing, contents path checks, alias resolution, list/install/resolve behavior, source-boundary scan, script smoke tests, realistic prompts, and negative prompts.
- Produce the blueprint artifact.
- Use
templates/role-blueprint.mdfor human review. - Use
schemas/role-blueprint.schema.jsonwhen the blueprint should be machine-checked.
- Use
Output
Return:
- blueprint summary;
- single-role/multi-role/topology decision;
- contents map and exclusions;
- permission and adapter rationale;
- validation plan;
- explicit write scope and stop conditions.
04 同仓库更多 Skills
example-skill
暂无简介
su-status
从 docs 人读真相源与 .ccb 协调件读取当前项目、Requirement 或子任务状态。
su-revise-breakdown
根据用户反馈重写 Requirement breakdown draft。
su-review
审查执行回执、diff 和验证证据,决定通过、返工、replan 或升级用户。
su-resume
从 docs 人读真相源与 .ccb 协调件恢复 paused/deferred/in-progress 主体的上下文。
su-reconcile
AI-native 自检 docs 人读真相源、.ccb 协调件与 Console 投影漂移,并按审批结果执行修复。