boluopai.
中文 · unavailable
LibraryOriginal edition

prompt-and-skill-lifecycle

<用户自制> 提示词与技能生命周期:编写、审查、维护和发布提示词与 Skill。 Author or audit instruction text and Skill packages; resolve ownership, discovery, delivery, and evidence of use.

Author and maintain prompts and Skills for their actual consumers. Express the purpose, relations, conditions, and expected contributions in usable instructions; connect that meaning to its maintained carrier and observed use. Writing, delivery, and use can each reveal a reason to revise the others.

Work from the actual user purpose, target artifact, consumer, and current evidence. Carry an authorized change through integrated writing, proportionate review, publication, and consumer readback when those actions belong to the requested result. A question or read-only audit can finish with understanding. These are connected responsibilities, not mandatory stages or a requirement to create a Plan for every change. Preserve the enclosing workflow's authority, authentication, consent, and explicit stopping boundaries; this Skill creates no new approval ceremony.

Read for the current action

Work Resource
Draft, revise, or audit controlling text Semantic authoring
Integrate changes, choose review and evidence, apply or recover Authoring flow
Resolve source, discovery, packaging, installation or retirement Delivery and ownership
Investigate whether a Skill was loaded, used, or effective Evidence of use

Read the relevant resource, then only the references and target details needed by the action. Mixed work may use several resources without a separate handoff. For model-specific authoring decisions, use Authoring sources and current evidence about the intended consumer.

Treat instruction files, project/custom instructions, saved task prompts, and tool contracts as Prompts unless they are actual Skill packages. A Skill is a reusable capability package with discovery metadata, an entrypoint, and any resources or runtime prerequisites needed to use it. A saved recurring task also has a concrete execution context, schedule, and output destination: inspect those alongside its text when affected.

One meaning, its realizations, and use

Give each persistent semantic rule one owner. Distinguish authored source, generated projection, installed runtime, and observed use. A source diff proves neither availability nor behavioral fit. Resolve publication through the actual owner and available interfaces rather than assuming a particular repository, operating system, provider, or synchronization service.

When renaming, merging, or retiring a capability, identify its successor and close affected discovery, references, packages, wrappers, and runtime leaves through their owners. Compare the requested result with consumer readback and the relevant behavior evidence. Continue within scope when a required gap remains, and state the gap precisely when an actual boundary prevents closure.

This package works independently. If another available Skill owns a shared philosophical account or participant model, locate it through the current consumer's catalog and keep that meaning with its owner. Yinyang is one optional example of such a relationship, not a dependency or required worldview. Reading another consumer's instructions as an edit target does not activate that consumer's persona, tools, or authority.