<p align="center">
<img src="assets/banner.svg" alt="Julius Brussee — Software Engineer. Building tools that respect your time." width="100%">
</p>
# Julius Skills
Eight personal agent skills for now: Caveman base, Interface Kit, Grill Me, Loop Factory, Junior to Senior, Deslopify, Context Canary, and The Last 20%.
This repo is shaped by eight things:
- **Caveman** - 70k-star token compression without technical loss. Small mouth, big brain.
- **Interface Kit** - accessible, performant interfaces with strong aesthetic direction, not generic AI slop.
- **Grill Me** - calibrated pressure before hard critique, so challenge matches user knowledge and comfort.
- **Loop Factory** - spec-driven agent loop where tasks move through inbox → active → archive with a real review gate.
- **Junior to Senior** - adversarial senior review that treats agent output as junior work and upgrades it with codebase + web research.
- **Deslopify** - mechanical scan-and-rewrite loop that erases AI-writing tells from any text and lands it in the right register.
- **Context Canary** - per-turn canary signal that makes silent context degradation visible, plus a recovery protocol when it trips.
- **The Last 20%** - finds and finishes the experiential layer agents skip: what the output reads like, which pages exist, first-run, defaults, the golden path.
Point is control. Agents should be terse when talking, precise when building interfaces, calibrated when challenging plans, and disciplined when running build loops.
## Quickstart
Install with `skills.sh`:
```bash
npx skills@latest add JuliusBrussee/skills
```
Pick the skills you want for Claude Code, Codex, Gemini, Cursor, Windsurf, Cline, Copilot, or any agent supported by the installer.
For the canonical Caveman installer, use:
```bash
curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/main/install.sh | bash
```
## Why These Skills Exist
Most agent output fails in three boring ways: too many words, UI that looks like every other generated demo, or critique that starts too hard before understanding user context.
This repo keeps three fixes close:
1. **Speak less, say more.** Caveman cuts output tokens while preserving exact commands, code, errors, and technical meaning.
2. **Build interfaces with taste.** Interface Kit starts from accessibility, performance, typography, spatial rhythm, color roles, and concrete interaction states.
3. **Challenge at the right altitude.** Grill Me assesses knowledge and desired pressure first, then asks one question at a time.
## Skills
### `caveman`
Ultra-compressed communication mode. Cuts token usage by dropping filler, hedging, pleasantries, and excess grammar while keeping technical accuracy intact.
Use when you want:
- shorter agent replies
- less token waste
- exact code, command, error, and API preservation
- persistent terse style until user exits with "normal mode"
### `interface-kit`
Implementation guide for high-quality UI. Synthesizes accessibility, performance, typography, layout, color systems, motion, interaction states, and component craft.
Use when building:
- frontend components
- landing pages
- dashboards
- design systems
- polished app flows
- accessibility fixes
- UI review passes
### `grill-me`
Calibrated interview skill for stress-testing plans, designs, and decisions. It first asks how much the user knows and how hard they want the pressure, then ramps from clarifying questions to failure-mode critique.
Use when you want:
- plan critique without getting overwhelmed
- one question at a time
- recommended answers with each question
- pressure matched to beginner, working, or expert knowledge
- softer or harder grilling on command
### `loop-factory`
Spec-driven agent loop. Coding tasks live as markdown specs that move through `inbox → active → archive`, get implemented by Claude Code or Codex, and must pass a review gate before they count as done. State is just which folder a spec is in. The governing rule: automate implementation and verification, not product decisions.
Use when you want:
- repeatable, reviewable agent work instead of one-off prompting
- visible task state (inbox / active / archive) with no dashboard
- generated implementation, review, and backprop prompts for either agent
- a hard review gate before anything is marked done
- to install or scaffold the `loop-factory` CLI into a project
Pairs with the [Loop-Factory](https://github.com/JuliusBrussee/Loop-Factory) repo, which ships the CLI and native Claude/Codex adapters.
### `junior-to-senior`
Adversarial review skill for agent-generated plans. Treats the current output as the work of a junior, then constructs a senior reviewer grounded in codebase research and web research of current best practices. Diagnoses altitude failures — plans that are foggy on the hard parts or tunneled into details with no product vision — and rewrites them into a scoped, state-of-the-art version with evidence behind every finding.
Use when you want:
- a staff-engineer-grade review of a plan before committing to it
- plans that commit on interfaces, versions, and failure modes instead of hand-waving
- best practices refreshed past the model's training cutoff via live web research
- a clear delta between the original plan and the upgraded one
- product decisions surfaced as open questions instead of silently invented
### `deslopify`
De-slop pass for any text. Scans with a regex catalog of AI-writing tells — negative parallelism ("not X but Y"), puffery vocabulary, rule-of-three, false ranges, em-dash abuse, uniform cadence, hedged both-sidesing — then rewrites at the level of meaning and re-scans its own output until clean. Built as a loop because the worst tells are emergent generative habits that a rewriting model reintroduces in paraphrase.
Use when you want:
- AI-drafted text that reads like a person wrote it
- a diagnosis table of which tells were found before the rewrite
- register-aware output: academic article, tweet, reddit post, LinkedIn, email, blog, docs, marketing
- no overcorrection — no fake typos, forced slang, or invented specifics
### `context-canary`
Early-warning system for long agent sessions. Installs a byte-stable first-line signal — the user's name, a turn counter, and an honest context self-check — so the moment the agent's hold on its instructions degrades (attention drift, compaction, truncation), the signal visibly dies. Comes with a trip protocol: checkpoint state to a file, re-anchor on project instructions, reset deliberately.
Use when you want:
- to know *when* a long session starts rotting instead of finding out from bad output
- a zero-infrastructure health check that runs every single turn
- compaction events surfaced the moment they happen
- a disciplined recovery path (checkpoint → re-anchor → fresh session) instead of limping on
Grounded in context-rot research (Chroma), lost-in-the-middle (Liu et al.), and instruction-drift findings — sources linked in the skill's references.
### `last-20-percent`
Finds and finishes the last 20% of a built solution — the experiential layer agents skip. Agents decompose the noun ("wiki" → ingestion, search, LLM) and the experience of using the thing never appears in that decomposition, so it never becomes a task. This skill decomposes the *scene* instead: one concrete magic moment, walked step by step, with the residue specced at plumbing fidelity and hand-crafted golden artifacts set as the quality bar before any generator gets built.
Use when you want:
- builds that end with a product, not a technically-complete demo
- the experiential work (content, IA, first-run, defaults, microcopy) specced as concretely as the plumbing
- golden artifacts written by hand before the machinery that generates them
- a final walk-through as the end user before anything is called done
- an audit mode for existing solutions that work but feel flat
## Interface Kit Standard
If a repo has `DESIGN.md`, it wins. Otherwise UI work should still have a point of view:
- Accessibility first: contrast, keyboard navigation, focus states, semantics.
- Performance before decoration: stable layout, lazy assets, transform-only motion.
- Typography matters: readable type scale, sane line length, tabular numbers for data.
- Layout uses stable spatial rules: 4/8px spacing, responsive constraints, no overlap.
- Components have complete states: hover, focus, active, disabled, loading, empty, error.
- Avoid generic defaults: no faceless hero, no purple-blue gradient template, no stock-looking polish.
## Caveman Ecosystem
This repo sits next to broader Julius agent stack:
| Repo | What |
|---|---|
| [caveman](https://github.com/JuliusBrussee/caveman) | Output compression for agents. |
| [caveman-code](https://github.com/JuliusBrussee/caveman-code) | Terminal coding agent built around token efficiency. |
| [cavemem](https://github.com/JuliusBrussee/cavemem) | Cross-agent memory. |
| [cavekit](https://github.com/JuliusBrussee/cavekit) | Spec-driven build loop. |
| [cavegemma](https://github.com/JuliusBrussee/finetune-caveman) | Fine-tuned model experiments for terse agent output. |
## Skill Reference
Run:
```bash
node scripts/list-skills.mjs
```
Verify frontmatter:
```bash
node scripts/verify-skills.mjs
```
## Contributing
Skills should be:
- Small enough to read quickly.
- Triggered by clear user intent.
- Progressive: load only the extra files needed.
- Specific about workflow and validation.
- Free of repo-specific assumptions unless the skill says so.
## License
MIT.