convention-engineering
Skill0 starsChanged 6 months agoMITView on GitHub
Use when designing or auditing repo conventions: `.conventions.yaml` descriptor, agent contract files (CLAUDE.md / AGENTS.md), docs taxonomy, verification gates, Git workflow, repo-local skill placement, `.work` layout, `.wiki` rules. Pick an archetype (whole-repo shape: epic-wrapper for multi-repo products, brain for personal/knowledge-base repos) and stack recipes (core-beliefs, docs-taxonomy, agent-knowledge, git-workflow, work, wiki, project-skill-placement, verify-script). Stack-agnostic. Skip for one-off product naming or domain architecture questions where no convention surface is being changed.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- attack-architectureSkill
- go-scriptingSkill
- harness-routerSkill
- paper-vettingSkill
- skill-builderSkill
- skill-cli-authoringSkill
- slock-epic-protocolSkill
- taskfile-authoringSkill
- work-cliSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 66
Read or create .conventions.yaml at the repo root.
- Distinctive: line 74
No auto-fixes without user approval.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Convention Engineering
Repo conventions for AI-agent-operated codebases. Stack-agnostic, descriptor-driven.
The pattern: a single .conventions.yaml at the repo root declares which conventions the repo opts into. An agent reads it, scaffolds or audits against it, and the same file feeds task verify.
Three building blocks
- Concepts — principles and the descriptor reference. Always-on.
- Archetypes — whole-repo shapes; pick one. (
epic-wrapper,brain.) - Recipes — stackable conventions; layer many. (
core-beliefs,agent-knowledge,docs-taxonomy,git-workflow,project-skill-placement,verify-script,work,wiki.)
A repo declares its archetype + recipes in .conventions.yaml, then runs task verify against the resulting opt-ins.
Routing
Concepts
| Question | Reference |
| What invariants does this skill optimise for? | references/concepts/agent-first-principles.md |
What goes in .conventions.yaml? | references/concepts/conventions-yaml.md |
| How should the verify gate work? | references/concepts/verification.md |
Archetypes
| Whole-repo shape | Reference |
Multi-repo product with repo/<leaf> symlinks | references/archetypes/epic-wrapper/pattern.md |
| Personal/knowledge-base brain repo | references/archetypes/brain/pattern.md |
| Adding a new archetype — format spec | references/archetypes/_format.md |
Recipes
| Stackable convention | Reference |
| Core beliefs / invariants / anti-patterns trio | references/recipes/core-beliefs/pattern.md |
CLAUDE.md / AGENTS.md structure | references/recipes/agent-knowledge/pattern.md |
Docs taxonomy (docs/{requests,planning,plans,…}) | references/recipes/docs-taxonomy/pattern.md |
| Execution-plan lifecycle | references/recipes/docs-taxonomy/exec-plans.md |
| Git commit, verify, push, and dirty-tree workflow | references/recipes/git-workflow/pattern.md |
| Repo-local skill placement | references/recipes/project-skill-placement/pattern.md |
scripts/verify.sh shape | references/recipes/verify-script/pattern.md |
.work/ tracker adoption | references/recipes/work/pattern.md |
.wiki/ knowledge base adoption | references/recipes/wiki/pattern.md |
| Adding a new recipe — format spec | references/recipes/_format.md |
Meta (about the skill)
| Concern | Reference |
| Bootstrap a new repo (generic flow) | references/meta/bootstrap-workflow.md |
| Audit an existing repo (generic flow) | references/meta/audit-workflow.md |
| Score a repo against its declared conventions | references/meta/evaluation-rubric.md |
| Evaluation report layout | references/meta/evaluation-report-template.md |
| Adapter contract for harness mirroring | references/meta/adapter-contract.md |
Quick Start
- Read or create
.conventions.yamlat the repo root. - For a new repo:
references/meta/bootstrap-workflow.md. - For an existing repo:
references/meta/audit-workflow.md.
Boundaries
- This skill prescribes structure, not stack-specific tooling. Linters, type checkers, supply-chain policies are the repo's choice; declare them in the descriptor and run them via
task verify. - The audit produces a gap report ("X is missing"); the evaluation rubric produces a graded judgment ("legibility 3/4, here is why"). Both live in
references/meta/. - No auto-fixes without user approval.
- For path migrations from the 1.x layout, see
MIGRATION.md.
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.