oberskills

oberskills

Six commands that make Claude Code better at the things it's worst at: writing like a person, searching the web without hallucinating URLs, building skills that actually work, and not embarrassing itself when dispatching agents.

Commands

write

Two modes. EDIT rewrites silently. REVIEW walks you through issues one batch at a time, asks questions, then offers an edit pass.

The skill loads in three layers depending on what the piece needs:

Layer 1 — Core (always loaded, ~160 lines)
  Five structural rules that won a blind test against a 3-judge panel:
  lurch (sentence length variance), spike (density variation),
  wander (non-linear structure), shift register, get specific.

Layer 2 — Surface (auto-loaded for EDIT, ~170 lines)
  Kill list, em-dash ban, hollow openers, hedge limit,
  transition crutches, contractions, sycophancy, verb poverty.

Layer 3 — Deep Craft (on demand, ~200 lines)
  Syntactic variety, rhetoric calibration, discourse templating,
  register range, name patterns, cliche metaphors, parallelism.

Typical edit: ~330 lines of instruction. Deep edit: ~530. Built on 47 AI-writing detection papers, Pangram Labs data (N=millions), and a blind test that dropped AI detection probability from 85% to 15%.

web-research

Parallel search agents fan out across multiple dimensions (docs, tutorials, discussions, forums). Each agent extracts precise information with source URLs. Results synthesize back through your model. No hallucinated links.

skill-craft

Creates and reviews Claude Code skills. CREATE mode runs intake, design, build, test (baseline/compliance/loopholes), and ship. REVIEW mode loads a checklist and audits structure, efficiency, and security. Both produce concrete artifacts, not advice.

prompt

Prompt engineering principles. Invoked standalone or chained through the agent command. Covers constraints, validation, and the structural patterns that make prompts stick.

agent

Validation gate before dispatching subagents. Chains into prompt engineering. Catches the things that go wrong when you hand work to a subprocess without thinking about it first.

shot

Screenshot capture and analysis. Full screen, active window, or named window. Dispatches a haiku-tier analyzer and returns a summary.

How They Connect

web-research ─── parallel search agents ─── synthesize with source URLs

skill-craft ──┬─ CREATE: intake → design → build → test → ship
              └─ REVIEW: checklist → audit → verdict

agent ──── prompt (constraints, validation)

write ──┬─ EDIT: core rules + surface rules (+ deep craft if needed)
        └─ REVIEW: scan → orient → top issues → next batch → offer edit

shot ──── capture → haiku analyzer → summary

Install

/plugin marketplace add ryanthedev/rtd-claude-inn
/plugin install oberskills@rtd
/plugin update oberskills@rtd

Version

1.27.0

---

MIT