Analysis revision 84fdeffGitHub snapshot Aug 12, 2026Metrics are a dated GitHub snapshot and are not used as a quality score.
This collection deliberately refuses to become one universal lifecycle. Explicit orchestrators compose work; automatically selected Skills contribute one engineering discipline; repository setup keeps both aligned with local conventions.
User-invoked Skills such as grill-with-docs and implement orchestrate a flow; model-invoked Skills such as diagnosis, TDD, research, and code review contribute one reusable discipline. A repository setup step first records the issue tracker, triage labels, and domain-context or ADR locations that later workflows must follow.
Three responsibilities that must not blur together
The most important design choice is the invocation boundary. It prevents an automatically triggered helper from silently taking over an entire engineering process.
Orchestrators
grill-with-docs, to-spec, and implement own an explicit flow requested by the user.
Disciplines
Diagnosis, TDD, research, domain modeling, and review each supply one reusable method.
Setup contracts
Tracker choice, triage vocabulary, and CONTEXT.md or ADR locations are recorded before later workflows depend on them.
Engineering ideas worth borrowing independently
Domain language becomes repository state
grill-with-docs sharpens terms in CONTEXT.md and records consequential decisions in ADRs so later sessions share the codebase’s nouns.
Diagnosis starts with a red-capable command
The diagnosing Skill refuses to theorize until one unattended command reproduces the exact symptom, then minimizes it and tests falsifiable hypotheses.
Architecture uses a precise module vocabulary
codebase-design looks for deep modules—substantial behavior behind a small interface—at explicit, testable seams instead of proposing broad rewrites.
Start with repository setup, then pick one discipline
Choose either the managed Claude plugin or editable skills.sh installation; installing both duplicates the skills.
Run the repository setup once so issue tracking, labels, and domain-document locations are explicit.
For a difficult bug, make the exact symptom fail from one fast command before ranking three to five falsifiable hypotheses.
Choose modular control over a single lifecycle
Strong fit for
Repositories that need better specification, domain modeling, work breakdown, TDD, difficult bug diagnosis, or code review.
Setup cost
The engineering workflows expect repository-level setup. Without tracker, label, and documentation conventions, their defaults can drift from the project.
Sources reviewed
84fdeff