dev-skills

Autonomous dev workflow, one plugin.

32 skills and 4 sub-agents for Claude Code — design, plan, implement, test, review, and release, each with a lite/full/ultra delivery mode.

View on GitHub Browse skills
/plugin marketplace add amjadjibon/skills

Pipeline

Skills compose end to end, or run standalone.

dev-loop
  └─ dev-research  →  dev-design   →  dev-create-plan  →  dev-review-plan  →  dev-implement-plan  →  dev-qa  →  dev-code-review
     (optional)         (optional)                                                    ↑                                 │
                                                                                       └──── fix agents ─────────────────┘

dev-design's API axis     → dev-api-design (+ openapi-spec for the spec doc)
dev-implement-plan phases → dev-tdd (test-first), dev-e2e-testing, dev-smoke-testing (on demand)

Skills

25 behavior overlays, invoked via the Skill tool.

dev-research

Investigate a codebase, approach, or technology before planning; writes RESEARCH.md with a recommendation.

dev-design

Decide a feature's shape: system design, data model, API contract, and UI/UX. Writes DESIGN.md.

dev-api-design

REST/GraphQL design principles: resource/URL design, pagination, versioning, GraphQL schema-first.

dev-ui-design

Builds a clickable UI prototype as one self-contained HTML file. Writes prototype.html.

dev-create-plan

Writes PLAN.md with phased, checkbox-driven steps ready for autonomous execution.

dev-review-plan

Reviews a PLAN.md before implementation for vague tasks, missing criteria, and scope issues.

dev-implement-plan

Executes a PLAN.md, ticking checkboxes and committing each phase.

dev-tdd

The red → green TDD loop: seams, anti-patterns, and the rules of the cycle.

dev-e2e-testing

Playwright end-to-end tests: what earns one, fixtures, and treating flake as a bug.

dev-smoke-testing

One-check-per-file sanity scripts run on demand — is the build alive, not a pipeline gate.

dev-qa

Measures test coverage, identifies untested paths, writes missing tests, produces QA.md.

dev-code-review

Reviews a diff or branch for correctness, security, and simplification; writes REVIEW.md.

dev-loop

Orchestrates plan → implement → review autonomously, spawning parallel agents in worktrees.

dev-refactor

Restructures code without changing behavior, applying changes in small verifiable steps.

dev-debug

Reproduce → isolate → fix → verify loop, committing the failing test before the fix.

dev-perf

Measure baseline → profile → optimize one bottleneck at a time → benchmark.

dev-clean-up

Removes merged branches, prunes stale refs, closes resolved issues, cleans worktrees.

dev-release

Derives the version bump from conventional commits, generates a changelog, tags a release.

openapi-spec

Writes and validates OpenAPI 3.1 spec documents — $ref components, Spectral/Redocly lint rules.

mermaid-diagram

Generates Mermaid diagrams with high-contrast styling and syntax validation.

github-actions

Creates and reviews GitHub Actions workflows, matrix builds, and security hardening.

git-commit

Writes conventional commit messages — type/scope selection, breaking-change marker, body and footer rules.

git-safe

Safety gate for destructive git ops, plus this repo's canonical commit hygiene.

brainstorming

Interactive ideation partner that pushes back instead of just agreeing.

prototype

Throwaway code in the real codebase to answer one design question — a driveable TUI, or UI variants on a real route.

Sub-agents

Parallel workers the skills spawn, each in an isolated worktree.

dev-researcher

Answers one scoped research question.

dev-implementer

Implements one PLAN.md phase.

dev-fixer

Fixes a group of REVIEW.md findings.

dev-tester

Writes coverage for one module's gaps.

Status line

Repo, model and context on one line; what the session has cost you on the next.

The dev-skills status line, two lines. First: [dev-skills] skills(main) · Opus 5 medium · ctx 324k/1M 32% · statusline polish · 45fdd1e0-af3f-46b4-be2e-9f8f802398a0. Second: usage: 5h 40% [resets in 3h53m] · 7d 20% [resets in 3d9h] · cost: ~$71.99 · time: 13h34m[api 1h11m] · edits: +1098/-259

Line two names each group and brackets any figure that qualifies another, and each percentage shares one traffic light — green under 50%, yellow under 80%, red at 80% and above, in muted hues chosen for distinct colour rather than intensity. The rate-limit segments show how much of each window is spent and when it refills (usage: 5h 40% [resets in 3h53m]); there is no credit balance in the session payload, so those windows are the only real quota signal, and line two is ordered by urgency so a narrow pane drops the notional cost before them. Segments drop out when the data isn't there yet, so a fresh session shows just the first line — and again when the terminal is too narrow, so neither line ever wraps.

A SessionStart hook wires it into settings.json on the first session after install, because plugins can't declare a statusLine themselves. It never overwrites a status line you already had, never replaces a refreshInterval you chose, and if you delete the one it installed it stays gone — each guarantee has a test. Opt out up front with touch ~/.claude/.dev-skills.statusline-optout.