Skip to content

JSRossie/agent-standards

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Standards

Reusable, cross-tool rules and skills that apply to every project. Behaviors are defined once here and consumed by multiple agent tools — Claude Code via skills, other agents (e.g. GitHub Copilot) via web-consumable renders.

Renamed from claude-standards — the behaviors aren't Claude-specific. See dotClaude ADR-0014 for the cross-tool model (two-tier topology, dual rendering).

Layout

  • baseline.md — the always-on portable nucleus: rules that apply to every project in any environment (commit format & posture, hard rules, concurrent-git hygiene). Designed to be imported whole into a project's CLAUDE.md without ever contradicting a sane project.
  • fragments/opt-in import fragments for environment- or project-specific machinery a project references by name only if it applies. Imported the same way as baseline.md. See Baseline & fragments below.
  • skills/ — Claude-native skill source folders (the canonical definition of each behavior).
  • github/ — web-consumable renders of those skills (AGENTS.md + optional *.prompt.md) for agents that don't read Claude Code skills. See github/README.md.

baseline.md (always) vs fragments/*.md (opt-in) is the core split: baseline carries only what's true everywhere, so importing it never pulls in dead weight or guidance that fights the importing project. Decision recorded in dotClaude ADR-0016 (the split) on top of ADR-0014 (the cross-tool hub).

Using a behavior in a new project

Claude Code — reference the skill

Skills install globally via dotClaude's .chezmoiexternal (pulled into ~/.claude/skills/), or copy a skill folder into a project. The project's CLAUDE.md then references it by name.

Other agents (e.g. Copilot) — copy the render

Clone this repo and copy the relevant github/<behavior>/ render into your target repo: AGENTS.md at the repo root (merge if one exists) and any *.prompt.md into .github/prompts/. Commit them there so they travel with the repo.

Baseline & fragments

baseline.md is imported into a project's CLAUDE.md. In environments that honor @<path> imports (Claude Code), reference it directly:

@~/Local/agent-standards/baseline.md

Environments that don't expand imports (e.g. Cowork) inline the content between sync markers instead:

<!-- BEGIN: agent-standards/baseline.md -->
[paste current contents of ~/Local/agent-standards/baseline.md, stripping its H1]
<!-- END: agent-standards/baseline.md -->

Refresh by replacing everything between the markers with the latest file. A small sync-baseline.sh (find the markers, splice in the latest file) is a reasonable addition once there are more than two consumers.

Fragments are imported the same way@~/Local/agent-standards/fragments/<name>.md in Claude Code, or inlined between their own <!-- BEGIN/END: agent-standards/fragments/<name>.md --> markers in Cowork — but only by projects they apply to. A project picks the fragments it needs:

Fragment Import when
fragments/commit-review-helper.md the project has a commit_review.py-style auto-split commit helper
fragments/cowork-sandbox.md the project runs in a Cowork sandbox (file-delete recovery, gc.auto=0, debris handling)

Distribution

This is a public git repo (JSRossie/agent-standards). Claude Code skills are pulled into ~/.claude/skills/ via dotClaude's .chezmoiexternal (weekly, or chezmoi apply --refresh-externals); the handoff distribution decision is recorded in dotClaude ADR-0012. GitHub renders are copied per-repo (see above). Forked/inlined copies stay frozen until manually re-synced.

Shipped

Behavior Skill (Claude-native) Render (cross-tool)
citation-standards skills/citation-standards/
handoff-protocol skills/handoff-protocol/ github/handoff/
planning-protocol skills/planning-protocol/ github/planning/

handoff-protocol is parameterized on handoffs_root / git_posture per project; derived from VRT. See dotClaude ADR-0012 (the standard) and ADR-0014 (the cross-tool render).

planning-protocol is the planning-weight ladder (plan mode → mini-spec → briefing/ADR → full spec toolkit), parameterized on decisions_root / briefings_root / max_rung / status_doc per project; derived from r-net. See dotClaude ADR-0021. The worklog/status machinery and lane model that surround it in r-net are deliberately not extracted yet — the ADR records the phase triggers for both.

Not extracted yet

Remaining portable concerns (document-conventions, document-timestamping, research-citation-schema) get pulled from VRT on demand as new projects need them. (commit-review-helper is now extracted — see fragments/.)

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors