skills

Connected skills that carry work from fuzzy plan to landed branch, on any project. Not a pile of skills; one flow with a fast path and a deep dial.

Install

Two paths, both for Claude Code. Pick one: installing both loads every skill twice.

Plugin (updates via version):

claude plugin marketplace add pwguler/skills
claude plugin install pwguler-skills

Or in a Claude Code session:

/plugin marketplace add pwguler/skills
/plugin install pwguler-skills

npx skills:

npx skills add pwguler/skills

Restart Claude Code after installing.

Update

Plugin:

claude plugin update pwguler-skills

npx skills:

npx skills update

Restart Claude Code after updating.

The loop

The base suite (drill → implement → verify → land) implements the run-until-done loop: a bounded goal, a maker/checker cycle, an exit only on proven criteria. Fast path: an obvious task skips the interview; one sentence is the spec, once it passes the fork test. If verify fails twice, the task was lying: route back through drill. /deep is user-only and turns every fast path off for the session.

flowchart TD
    idea([idea or codebase friction]) --> drill["1. drill: settle the plan"]
    drill -- "goal, non-goals, acceptance criteria" --> spec[("docs/specs/*.md")]
    drill -- "tree too big for one session" --> draft[("spec as draft: open decisions")]
    draft -- "sessions resolve decisions one at a time" --> spec
    spec --> impl

    subgraph cycle["2. the maker / checker cycle"]
        impl["implement (maker): failing test, minimum code"] --> verify["verify (checker): reject by default, quotes evidence"]
        verify -- "criterion fails" --> impl
        impl -. "bug: root cause first" .-> debug
        debug -.-> impl
    end

    verify -- "every criterion proven" --> land["3. land: fresh-eyes gate"]
    impl -- "3 failed attempts on one slice" --> you(["escalate to you"])
    land -- "residue to ADRs, CONTEXT.md, ARCHITECTURE.md; spec kept unless dropped" --> done([branch closed])

Loading

The documents the suite maintains, all created lazily with no setup step: CONTEXT.md (domain glossary, born from the first resolved term), ARCHITECTURE.md (the system as it is), docs/adr/ (hard-to-reverse decisions), docs/specs/<slug>.md (the loop's steering artifact; a spec kept after its branch is a record of why the code looks the way it does).

When to use what

Moment	Skill
New plan, feature, or design	drill
Implementing a settled plan	implement
Bug or unexpected behavior	debug
About to claim done, fixed, or passing	verify
Existing code fights you	architecture
Reading legwork	research
Branch done, needs merging or a PR	land
Independent failures or tasks, two or more	parallel
Work that needs its own full session, in a tmux pane	spawn
Multi-task plan to execute hands-off	parallel
Authoring or editing anything an agent reads	write-skill
Learning a topic across sessions	teach
A message did not land, you stopped following	huh
Risky or ambiguous work, full rigor wanted	deep
None of the above comes to mind	which-skill

core-interview is internal support: drill and architecture run it, you do not invoke it directly.