Project Instruction Builder
Turn conversation patterns and preferences into a durable project instruction file.
The skill requires reviewing conversation context, extracting user preferences, and clarifying scope when rules are unclear. It requests no credentials, network access, or inherently high-risk action. However, saving the file lacks explicit pre-write confirmation, scope controls, rollback guidance, data-flow disclosure, and sensitive-data handling, so points are deducted.
The workflow is internally coherent: clarify, draft, iterate, and summarize. However, it depends on the unavailable agent-customization guidance, does not specify the target path or validation format, and provides no handling or diagnostic feedback for write failures or abnormal inputs. No skill-specific tests are supplied, so the static score is limited.
The name, description, and argument hint clearly target creation of project-rule instruction files, while disabled model invocation reduces accidental triggering. Boundaries for non-fit cases, input/output constraints, varied project layouts, Chinese-language support, and environment limitations are not stated, limiting trigger precision and applicability evidence.
The YAML metadata, purpose, and main workflow are readable and somewhat layered. However, the key related guidance is an unprovided dependency, and the skill lacks installation notes, concrete file examples, FAQs, known limitations, versioning, changelog information, and a clear maintenance path, so points are deducted.
The skill covers the core flow of extracting rules from conversation, clarifying scope, drafting the file, iterating, and summarizing the result. Yet it lacks a concrete template, validation criteria, and directly usable examples; it also requires user review. Static evidence cannot establish completeness or meaningful advantage over manual editing, so the score remains limited.
The skill text is auditable, and the repository supplies general CI, licensing, and governance material. There are no dedicated tests, execution records, or third-party reproduction artifacts covering this skill’s key paths, and nothing was executed in this review, so verifiability is limited.
- The skill depends on the unprovided agent-customization guidance, so its template and principles cannot be fully verified.
- Before saving, confirm the target path, content, and scope, and provide an undo or recovery path.
- Do not persist credentials, personal information, or other sensitive conversation content into an instructions.md file without filtering.
- No dedicated tests or execution evidence are provided; manually inspect the generated file’s format, scope, and resulting behavior.
What does this skill do, and when should you use it?
This skill creates a .instructions.md file for a project rule or convention. It reviews the conversation for repeated corrections, preferences, coding patterns, and project-specific conventions. When no clear rule is available, it asks about scope, affected technologies or file types, and whether the rule is mandatory or preferential; it may also explore the codebase. It then drafts and saves the instruction, asks about its weakest or most ambiguous parts, and summarizes the finalized rule with example prompts and related customization ideas.
Reviews the available conversation history; extracts user corrections, preferences, coding patterns, and project conventions; asks clarifying questions about scope, technologies or file types, and rule strength when needed; explores the codebase when more context is required; drafts and saves a .instructions.md file; identifies ambiguous or weak parts and requests feedback; and, after finalization, summarizes what the instruction enforces, suggests example prompts, and proposes related customizations. It also directs the agent to load and follow the instructions.md file from the related agent-customization skill, but that file is not included in the supplied material.
- A developer repeatedly corrects generated code and wants those corrections captured as a persistent project rule.
- A team needs a convention for a particular technology or file type but has not yet defined its scope.
- A user wants codebase context considered before turning informal project habits into an instruction.
- A user wants to review a draft, resolve its ambiguities through questions, and finalize the instruction iteratively.
What are this skill's strengths and limitations?
- Converts ongoing conversational feedback, coding patterns, and project conventions into a .instructions.md file.
- Clarifies scope, affected technologies, file types, and rule strength when the request is underspecified.
- Provides a draft, ambiguity review, feedback loop, final summary, examples, and follow-up customization suggestions.
- The repository README states that the source code is available under the MIT license.
- It depends on the related agent-customization skill and its instructions.md file, whose template and principles are not included here.
- The source does not specify the exact directory where the generated .instructions.md file is saved.
- The SKILL.md sets disable-model-invocation: true, so it should not be expected to run through automatic model invocation; it requires explicit user triggering or compatible client handling.
- No test suite, platform test results, or detailed failure behavior are provided for this individual skill.
How do you install this skill?
The supplied material does not document a dedicated installation command or installation directory. The skill is located at extensions/copilot/assets/prompts/skills/create-instructions/ in the microsoft/vscode repository, whose source README states that the repository is licensed under MIT. To use it, expose that skill directory to a compatible Agent Skill client and make the related agent-customization/instructions.md available.
How do you use this skill?
In a client where the skill is available, start with a concrete rule or convention, for example: “Create an instruction for this project: use single quotes in all TypeScript files, and apply the rule only under src/.” If the request does not establish a clear rule, the skill will ask about scope, technologies or file types, and whether the rule is mandatory or a preference. The source does not document a client-specific trigger syntax or an exact output location.