By dromsak
Personal toolbox of Claude Code skills. Headline workflow is the AFK pipeline (afk-issues → afk-army → uat) for backlog-driven parallel agent development; ships alongside architect, grill-me, and dromsak-guidelines.
Drain the `ready-for-agent` GitHub backlog by spawning one worker subagent per unblocked issue via the Workflow runtime — each implements in its own managed worktree, pushes, and opens a PR — then land them with a single batch-verify merge-train (integrate all, gate once, bisect on red). The orchestrator (this conversation) picks the queue and runs the merge, then LOOPS: after each merge-train it re-picks issues freshly unblocked by what just merged and drains the next wave, continuing until the reachable dependency graph is exhausted. The runtime owns worktrees, parallelism, and result tracking. Interactive, not unattended.
Carve findings from the current conversation into well-shaped GitHub issues. Batches by module by default — one issue per module carrying ALL that module's findings (worker context-loading dominates cost, so N findings per file-read beats N file-reads). Generates title + acceptance-criteria body, classifies complexity (needs-sonnet vs needs-opus), wires same-surface `## Blocked by` markers automatically, and tags `ready-for-agent`. Companion to `/afk-army`, which then drains the resulting queue.
Senior-architect audit of a code surface — finds bespoke wheel-reinvention, over-engineering, duplication, dead code, layering drift, and bloat, adversarially verifies each finding, ranks them, and (only on your pick) routes fixes through /afk-issues → /afk-army. Language-agnostic: auto-detects the stack and reads an optional per-repo `.claude/architect.md` profile for named surfaces and project rules. Read-first, never autonomous. Invoke as `/architect <scope>` where scope is a named surface (from the repo profile) or any path/file/fuzzy module name.
Senior design-craft audit of a *rendered* UI surface — drives the live browser to measure what a glance cannot (column voids, redundant chrome decks, stray borders, density drift, off-scale tokens, contrast), diffs sibling pages against each other, and reads every label as a first-time user, then adversarially verifies each finding against the measured numbers, ranks them, and (only on your pick) routes fixes through /afk-issues → /afk-army. Needs the dev stack up + Playwright. Reads an optional per-repo `.claude/design-audit.md` profile for the surface map, brand law, and deliberate-decision exceptions. The rendered-output twin of /architect (which audits source statically). Read-first, never autonomous. Invoke as `/design-audit <surface>`.
dromsak's operating rails for Claude Code — lean context engineering ("token-maxxing"), frugal delegation, deliberate verification, karpathy coding rails, advisory-lane autonomy. Load at the start of a session.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Personal toolbox of Claude Code skills, packaged as a plugin. The headline workflow is the AFK pipeline — a way to drive a GitHub-Issues-shaped backlog through a fleet of parallel Claude Code subagents without leaving the room mentally, but also without leaving the room physically (this isn't unattended overnight automation; it's "I can do other things in this conversation while five workers grind").
| Skill | What it does |
|---|---|
afk-issues | Carve findings from the current conversation into well-shaped GitHub issues with acceptance criteria, complexity labels, and same-surface ## Blocked by markers. |
afk-army | Drain the ready-for-agent queue via the Workflow runtime — one worker subagent per unblocked issue, each in a runtime-managed git worktree (implement → push → PR), then a single batch-verify merge-train lands them (integrate all, gate once, bisect on red). Pairs SKILL.md (orchestration) with afk-workflow.js (the parallel drain). |
uat | Drive Playwright-based UAT journeys against a running web app, filing GitHub issues inline. Self-bootstraps docs/uat.md on first run by interviewing you and reading your routes. |
architect | Senior-architect audit of one code surface — finds wheel-reinvention, over-engineering, duplication, dead code, layering drift, and bloat across 15 named lenses, adversarially verifies each finding, ranks by impact÷effort, and (only on your pick) routes fixes into the AFK pipeline. Language-agnostic engine: auto-detects the stack and reads an optional per-repo .claude/architect.md profile for named surfaces and project rules. Read-first, two human gates, no autonomous mode. Pairs SKILL.md with audit-workflow.js (the frugal ~2-agents/chunk fan-out). |
grill-me | Interview the user relentlessly about a plan or design until reaching shared understanding. Vendored from mattpocock/skills. |
dromsak-guidelines | dromsak's personal operating profile, centred on context engineering / token-maxxing — small relevant working set, just-in-time loading, parallel read-only subagents for fan-out, compact at theme boundaries, frugal delegation, verify-once-at-the-boundary. Its §4 consolidates the karpathy LLM-coding rails (MIT — see Attribution), and it adds an advisory-lane autonomy stance. Loaded by afk-army workers; load at session start. |
The three workflow skills (afk-issues, afk-army, uat) are designed
to work as a tight pipeline. They share a body template, a label
vocabulary (ready-for-agent, needs-sonnet, needs-opus,
agent-working, needs-human-review, needs-info), a branch prefix
(army/*), and a ## Blocked by convention for explicit issue
dependencies. (Same-surface serialization is no longer pre-computed —
file conflicts now surface and resolve once, at the merge-train's
integration step.)
Typical loop:
/uat — walk your Playwright journeys, file an issue per finding./afk-issues — carve any findings from the current conversation
that didn't come from /uat (code reviews, grilling sessions, bug
reports) into more issues./afk-army sonnet — fires one worker per unblocked issue (the
runtime paces concurrency), each implements + opens a PR, then the
batch-verify merge-train lands the green ones in the same run. No
separate merge phase — it pauses for you only if a worker escalates.You can use the skills independently, but mixing in custom labels or a different body template means hand-editing the skill prompts — the contract is hardcoded, not configurable.
This is a Claude Code plugin, served from the marketplace defined in
.claude-plugin/marketplace.json. Install is two steps — add the
marketplace, then install the plugin from it (you can't /plugin install
a bare repo URL):
/plugin marketplace add dromsak/skills
/plugin install dromsak-skills@dromsak-skills
(dromsak/skills is a public repo, so no auth is needed to add the
marketplace. To pick up new skills later: /plugin marketplace update dromsak-skills then reload.)
After install, the five skills are user-invocable as /afk-issues,
/afk-army, /uat, /grill-me, and /dromsak-guidelines.
afk-issues, afk-army, and uat all need:
npx claudepluginhub dromsak/skills --plugin dromsak-skillsReliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Design fluency for frontend development. 1 skill with 23 commands (/impeccable polish, /impeccable audit, /impeccable critique, etc.) and curated anti-pattern detection.
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls