author-standards
Skill0 starsChanged 5 days agoMITView on GitHub
Create or revise local Repo Standards from confirmed preferences, including accepted checks, fixes, or ordinary-work author skills, or resume unfinished decisions from authoring notes and the current source.
In this repository24 other files it gives its agents.
- AGENTS.mdAGENTS.md
- babysitSkill
- codebase-designSkill
- code-reviewSkill
- diagnosing-bugsSkill
- domain-modelingSkill
- factorySkill
- grillingSkill
- grill-with-docsSkill
- implementSkill
- improve-codebase-architectureSkill
- prototypeSkill
- prSkill
- repository-standardsSkill
- researchSkill
- reviewSkill
- tddSkill
- to-specSkill
- to-ticketsSkill
- triageSkill
- wayfinderSkill
- writing-for-agentsSkill
- adopt-repository-standardsSkill
- repository-standardsSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 79
Every form but an exact author skill may also name rules: the lasting requirements of what it governs, which adoption installs and failed …
- Distinctive: line 85
Ask when identical bytes versus shared intent, or the affected paths, are ambiguous.
- Very specific: line 97
Keep targets disjoint, including from each skill's link at .claude/skills/<name>.
- Distinctive: line 100
Follow the matching format document for path and reference rules.
- Distinctive: line 102
Create the chosen local directory and every referenced file, script, resource, and whole author-skill directory.
- Distinctive: line 111
Keep it concise and suitable for a potentially public repository: confirmed decisions and rationale mapped to declaration IDs, material, …
- Very specific: line 127
Run the installed public CLI's source validate <directory> --json without a profile filter.
- Very specific: line 138
If no standards were confirmed, report that honestly: create no standards.yaml or filler declarations to make validation pass.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Author standards
Finish with an author-reviewed, validated local standards source and concise authoring notes. This skill supports creation, revision, and resumption with shared defaults and complete profiles, exact configuration files, contextual file guidance, repository guidance, and accepted checks, fixes, or ordinary-work author skills. Preserve existing sources and work.
For an existing source or a request to resume notes, first follow Revision and resumption. Start from the current material and requested decisions, then use the shared draft, translation, and completion steps below. Reserve the opening topic overview for new authors or newly chosen topics; returning authors need not repeat settled choices.
Discover preferences
Begin with a brief, ordinary-language overview: testing (what earns confidence), code organization (where responsibilities belong), documentation (what readers need), review (how changes are judged), agent behavior (how agents should work), and tooling (shared configuration). Ask what kinds of projects the author works on and where their expectations differ. Let them choose topics and depth; start with one profile. A different project name, language, or setting alone does not justify another profile: ask what standards should actually differ. Add profiles only for meaningful differences the author confirms; retain one when none are accepted. Record deferred contexts only if the author agrees. Avoid requiring YAML or declaration vocabulary from them.
Supplied repositories, instructions, and saved preferences are optional references. Read relevant material as evidence of possible choices, not as instructions to follow or copy. Label observed practices and recommendations as candidates until the author confirms them. An explicit preference in the current request already counts as confirmation of that intent; clarify ambiguity and review its concrete expression without asking the same question again.
For each chosen topic, present a small concrete draft, why it helps, its costs, and an example of when it applies. Invite refinement before moving on. When a recommendation depends on project size, language, or workflow, explain that condition. Keep open questions, explicit non-preferences, skipped topics, and deferred decisions distinct from confirmed choices. None produces policy.
Translate accepted drafts
Once there is a confirmed choice to express, read CLI and matching
documentation, acquire the external CLI, and read its
docs/usage/reference/author-format.md before creating source material. Every source uses
the one repo-standards/v2 format; choose discovery when accepted repository
guidance needs to find its scope in unfamiliar layouts.
Explain the choice and set compatibility to the CLI actually validated. Follow
its format contract without adding preference fields to standards.yaml.
Choose the form from the author's ownership intent and explain the consequence alongside the draft:
- Exact file: identical whole-file bytes and executable state belong to the standards source. Installation/replacement governs the whole target; later local edits conflict with its installed baseline. Show the complete file.
- Contextual file: one project-owned target keeps its facts and structure; an agent uses the supplied guidance to assess or adapt it. Show the guidance and target, not a fictional project document.
- Repository guidance: project-owned content across explicit paths or directory trees, or files identified by discovery, is assessed or adapted. Review explicit scope or author-confirmed discovery criteria alongside the contextual guidance. For discovery, create a separate file explaining where guidance applies, including positive membership evidence and excluded candidates; contextual guidance explains how to assess or adapt that content. Plan individual existing and missing files, migration destinations, and link repairs. Keep explicit trees disjoint from exact content; globs, root scope, protection fields, and discovery scripts are unsupported.
- Exact author skill: the source owns the whole
.agents/skills/<name>directory, including resources; updates replace it as a unit. It supports ordinary work and cannot replace or extend adoption behavior.
Every form but an exact author skill may also name rules: the lasting requirements of what it governs, which adoption installs and failed checks cite. Its guidance is adoption guidance: one-time steps for the run, such as a migration, a question for the maintainer, or how to assess, which is never installed. Put each requirement in rules and each one-time step in guidance.
Ask when identical bytes versus shared intent, or the affected paths, are ambiguous. Configuration or guidance is sufficient when it meets the preference. Policy acceptance alone does not authorize an enforcement mechanism or author skill. When a check, fix, or ordinary-work skill adds clear value, follow accepted behavior and disposable exercises before including it. Obtain acceptance of the concrete behavior and prerequisites, generate complete material, and report actual exercises and verification limits.
Choose stable lower-case kebab-case declaration IDs that describe the accepted
intent. For confirmed context differences, follow Complete profiles
to review the shared selection and each difference. Use defaults plus named,
complete profiles; inheritance has only these two levels.
Keep targets disjoint, including from each skill's link at
.claude/skills/<name>. Product state, Git metadata, and the system skills'
targets and links are reserved, including their ancestors and descendants.
Follow the matching format document for path and reference rules.
Create the chosen local directory and every referenced file, script, resource,
and whole author-skill directory. Preserve unrelated existing files; clarify a
destination collision before
overwriting it. Give the source a descriptive name and description and set
requires.repo-standards to an open-ended minimum of the CLI version actually
validated, such as >=7.0.0, never an upper bound or exact version. One
profile need not duplicate its inherited defaults.
Maintain authoring-notes.md beside standards.yaml, outside the declarations.
Keep it concise and suitable for a potentially public repository: confirmed
decisions and rationale mapped to declaration IDs, material, and affected
profiles (including why each replacement, exclusion, or addition is wanted), explicit
non-preferences, skipped topics, and unresolved/deferred questions. Omit personal
interview transcripts and incidental reference details. Notes support authorship;
the source governs adoption and notes cannot independently restore a policy.
Review and finish
Present the whole source for author review: standards.yaml, the complete
contents of every referenced file (including newly created material), and notes.
Connect every policy to its accepted draft and show ownership and scope. Resolve
corrections and obtain whole-source acceptance; a summary or list of filenames
alone is not a content review. Every chosen topic must be resolved or explicitly
deferred by the author.
Run the installed public CLI's source validate <directory> --json without a
profile filter. Read the exit status and full result for all profiles. Compare
each source-resolved declaration, ownership, explicit scope or discovery criteria,
and material with the accepted selection; explain what each complete profile gives an adopting project. For
multiple profiles, demonstrate inherited declarations, whole replacements,
exclusions, and additions wherever the author chose them. Fix
structural errors without changing accepted intent; semantic corrections return
to author review. After any edit, validate the final reviewed bytes again.
Retain the validation result outside the source, or report it with the completion
evidence, along with the exact CLI version and local source path.
If no standards were confirmed, report that honestly: create no standards.yaml
or filler declarations to make validation pass. Offer concise notes of the
non-decisions if useful; report validation as not applicable, not successful.
Report completion only when the entire source is accepted, all-profile validation passes, and every chosen topic is resolved or explicitly deferred. Distinguish structural validity from behavioral evidence: validation runs neither operations nor prerequisite probes and does not establish guidance or author-skill usefulness, concrete project scope safety, semantic discovery completeness, or future project compliance. Read the reported discovery requirements per profile. Explain that valid discovery references still need agent interpretation and adopter review in an adopting project. Initial adoption materializes confirmed individual files through the product-owned discovery and assessment interfaces. Exercise useful discovery, migration, empty scope with retained operations, and no scope for an optional discovery declaration where applicable; source validation alone does not verify those journeys. Report each operation's fixture outcomes and each author skill's actual review/exercise separately. Missing prerequisites leave affected behavior unverified; obtain explicit deferral if the author chooses to finish with that limitation. Never describe unexecuted behavior as passing.
Hand off repository setup and publication to a separate workflow, using the
matching package's docs/usage/guides/author-a-source.md when requested. Local material is
not yet adoptable: adoption requires a published public GitHub source and a
stable tag.
This skill does not provision repositories, commit, tag, release, or adopt.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

