Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

brunodantas/skills

A collection of standalone Agent Skills for the SDLC loop, including:

  • Architecture
  • Verification
  • Review

Installation

Copy the skills into a project, under whatever agent directory it uses:

npx skills@latest add brunodantas/skills

Or install them as a Claude Code plugin, which keeps one copy and updates in place:

/plugin marketplace add brunodantas/skills
/plugin install sdlc@brunodantas

Plugin skills are namespaced, so they run as /sdlc:to-tests rather than /to-tests. To hand a whole team the collection, commit the marketplace to the project's .claude/settings.json and they are prompted to install it on first trust of the repo:

{
  "extraKnownMarketplaces": {
    "brunodantas": {
      "source": { "source": "github", "repo": "brunodantas/skills" }
    }
  },
  "enabledPlugins": { "sdlc@brunodantas": true }
}

Skills

Skill What it does
ilities Rank architectural characteristics at any scope, from a whole system down to one live decision, and offer to document the call.
to-cases Derive a spec's numbered ## Cases section from the behaviours it describes and a unit's acceptance criteria, and land it in a durable file.
to-tests Turn one unit of work into a frozen oracle: tests plus stub interfaces, offered as one freeze commit.
harden-tests Adversarially check tests against the cases they claim to encode, then write down what the pass could not close. Uses sub-agents where it can, and says so when it can't.
build-pr Write the branch's PR description as a PR Map: a review tier, evidence-backed claims, and every file bucketed read-hard / skim / skip. Best with GitHub and an authenticated gh; degrades to a body you paste.

to-casesto-testsharden-tests chain, but each runs on its own. Take to-cases alone if all you want is a spec convention that spells out its cases, or harden-tests alone to measure how weak an existing suite is.

No setup required

Every skill here runs with zero configuration: the formats are bundled, and anything a skill can't discover it asks you once. There is deliberately no setup skill to run first.

If you'd rather pin a convention than answer the question each time, drop a file at one of these paths and the skills will read it instead:

Path Read by What to put in it
docs/agents/cases.md to-cases Where specs live, and what your cases section is called if not ## Cases
docs/agents/testing.md to-tests, harden-tests The command that runs one test file, and where tests and fixtures live
docs/agents/review-tiers.md build-pr Your own tier trigger lists — start by copying REVIEW-TIERS.md

docs/review-tiers.md, .github/review-tiers.md, and any path your CLAUDE.md / AGENTS.md points at work too. docs/agents/ is just a common place to keep per-repo agent config; nothing here requires it.

Non-duplication policy

No skill here shadows the name of one in a widely installed collection, mattpocock/skills above all. If an idea from this collection is absorbed elsewhere, the skill here is deprecated with a pointer rather than maintained in parallel.

Status

0.2.0: five skills, installable as files or as a plugin. They are used daily in private repos, and the public collection is what that use has settled into. Expect the interfaces to keep moving through 0.x; see the changelog.

License

MIT. See LICENSE.

About

Agent skills for SDLC

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Contributors

Languages