From ck
Classifies dev tasks as quick, standard, or thorough via 5-axis rubric (files touched, type, judgment, cross-component, novelty) with overrides to set token budgets, model tiers, reviews, and tests.
npx claudepluginhub juliusbrussee/cavekitThis skill uses the workspace's default tool permissions.
A deterministic scoring rubric. Five axes, 0–4 each, summed to 0–20.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Searches prompts.chat for AI prompt templates by keyword or category, retrieves by ID with variable handling, and improves prompts via AI. Use for discovering or enhancing prompts.
Guides agent creation for Claude Code plugins with file templates, frontmatter specs (name, description, model), triggering examples, system prompts, and best practices.
A deterministic scoring rubric. Five axes, 0–4 each, summed to 0–20.
| Axis | 0 | 1 | 2 | 3 | 4 |
|---|---|---|---|---|---|
| Files touched | 0–2 | 3–5 | 6–10 | 11–20 | 20+ |
| Type | chore / format | refactor | feature | cross-cutting | architectural |
| Judgment required | mechanical | low-ambiguity | medium | high | critical (sec/prod) |
| Cross-component | single module | two modules | three modules | many within one repo | multi-repo |
| Novelty | known pattern | rare pattern | novel | research needed | unknown unknowns |
Total score maps to:
| Score | Depth |
|---|---|
| 0 – 6 | quick |
| 7 – 13 | standard |
| 14+ | thorough |
Upgrade one step regardless of score when any of these are true:
Downgrade one step only when all of these are true:
| Depth | Token budget | Model tier | Review | Tests |
|---|---|---|---|---|
| quick | 8 000 | haiku | optional | smoke |
| standard | 20 000 | sonnet | required | unit + integration |
| thorough | 45 000 | sonnet/opus | mandatory | unit + integration + E2E |
These defaults are recorded in .cavekit/config.json under task_budgets
and consumed by the cavekit-router.cjs model router.
The ck:complexity subagent (haiku) receives a task description and returns
a JSON blob:
{
"score": 11,
"depth": "standard",
"axes": {
"files": 2, "type": 2, "judgment": 2, "cross_component": 2, "novelty": 3
},
"overrides_applied": []
}
/ck:map calls this agent per task to set depth in the task registry. If
the agent produces a score in the "thorough" band with a novelty of 4 and a
security override, it may return needs_research: true, which /ck:map must
translate into an upstream ck:researcher task dependency before the work
itself.
/ck:sketch — runs complexity scoring on the whole domain to set the kit's
complexity: frontmatter./ck:map — runs it per task to assign depth./ck:make — reads depth to size the task budget and pick the review
intensity.ck:complexity agent — pure-haiku worker; does nothing else.