alvaromongon/claude-config

Versioned subset of ~/.claude: my engineering baseline and the skills that apply it. The repo
lives in place, so Claude Code loads it directly in every session.
At a glance
- **What it is**: a methodology, not an application — a personal engineering baseline
(common.md, everyday rules in
CLAUDE.md) plus skills that set up repositories against it and run a
GitHub-native loop — the owner's ideas are shaped and external reports triaged, then refined →
implemented → reviewed → merged, autonomously where a repo allows it.
- **Use it**: in any repo, /quality-baseline to set up or audit quality gates, then
/github-workflow-setup to enable the issue workflow. See Setting up a repository.
- **Quality**: documentation checks in CI (Docs workflow); behavioural
scenarios per workflow skill (skills/*/evals/); decisions in docs/decisions/.
Contents
| Path | What it is |
|---|---|
| CLAUDE.md | Global instructions loaded in every session: everyday engineering rules and working agreements. |
| skills/ /SKILL.md | One skill per folder; references/, templates/ and evals/ beside it. |
| docs/decisions/ | ADRs for this repository. |
| docs/credits.md | Prior art the GitHub workflow skills are modeled on. |
| docs/upstream-watch.md | How far check-upstream-skills has reviewed upstream. |
| docs/model-watch.md | When and how the per-skill model assignment is reviewed, and its log. |
| .github/ | Docs CI, its validation script, Dependabot and the ruleset to apply once public. |
Everything else in ~/.claude (history, sessions, caches, settings.json) is private runtime
state, excluded by the whitelist in .gitignore.
Skills
**Starts automatically** means Claude may invoke the skill on its own when the request matches its
description; every skill can also be started explicitly with / . **Manual only** skills
set disable-model-invocation: true: only typing the slash command starts them.
| Skill | Use it to | Starts | Model | Notes |
|---|---|---|---|---|
| quality-baseline | Set up or audit a repo against the baseline (tests, CI gates, pre-push, rulesets, Docker, ADRs…). | Automatically | sonnet | C# templates in templates/csharp/; language references in references/. |
| adr | Write, propose, supersede or list ADRs (MADR 4.0). | Automatically | opus | |
| github-workflow-setup | Set up or audit a repo's issue workflow; or get a reminder of the cycle. | Automatically | sonnet | Asks before widening any permission. |
| github-shape | Shape an idea into a spec issue with a success metric; resolve its decisions; close it checking the metric. | Automatically | opus | Starts from the owner's position; shows the draft before publishing. |
| github-triage | Classify incoming external issues and PRs. | Automatically | sonnet | Proposes; applies only after approval. |
| github-refine | Turn an agreed spec into a milestone of tracer-bullet issues. | Automatically | opus, effort high | Shows the draft before publishing. |
| github-implement | Implement one ready-for-agent issue with TDD, up to an open PR. | Automatically, e.g. /github-implement 42 | sonnet | Runs in a forked context; needs the issue number. |
| github-work-queue | Work through the ready-for-agent backlog unattended. | **Manual only** | sonnet; reviewers opus | Merges PRs where the repo allows it. |
| check-upstream-skills | Review mattpocock/skills for improvements; check the model assignment. | **Manual only** | session | Occasional, on demand. |
The built-in code-review and security-review skills are reused as the review gate of the work
queue.
**Model** is a family alias, so each skill follows that family's latest model; the assignment by
role and how it is kept current: ADR 0003
(github-shape, added later, is opus per ADR 0004),
docs/model-watch.md.