caveman
claudeTalk like caveman. Cut filler. Keep technical details explicit.
22 skills · 1 command · 3 agents · hooks
Commands, skills and hooks that do not need the jev plugin: explica-optiuni, compact-handoff, compact-session, class-graph, logic-test-debug, explica-clar, Android helpers, claude-auto, claude-openai, claude-copilot, explicit scope approval hooks, legacy claude-scope uninstall and jev-decision-loop (Jev over its direct API, the user's own key).
Plugin apana from marketplace apana-private in panaalexandrucristian/apana-plugin. Commands read from the repo's manifest — yesterday.
This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.
releasing-pluginReleases a new version of the apana plugin from this repository (version bump in 4 files, PRIVATE.md paragraph, release commit, tag, push and CI check). Use when the owner asks to release, ship, publish, tag or bump the version of the plugin, or says facem release.
android-ui-stepWhen you drive an Android emulator or device from the shell (open an app screen, tap a tab or button, check what a screen shows) with the `android` CLI or `adb`: use this skill's ui-step.mjs. ONE command taps a label, waits until the screen stops changing and prints a short one-line-per-element dige
assess-complexityDetermines the complexity of a plan or code change (refactor, migration, modularization, feature) BEFORE implementation, on two separate axes - size/blast radius (files, fan-in, modules, delicate code) and Cynefin domain (how well known the cause-effect relationship is) - and recommends the process
class-graphOn project-local Kotlin/Java code, with file:line evidence: (1) Kotlin function lines: where a named function starts/ends, its body (function body lines, then Read just that range), a file's function ranges; (2) who uses, calls or depends on a named class, interface, object, top-level function or me
claude-copilotUse when running Claude Code with a GitHub Copilot subscription, installing its standalone launcher, refreshing its model catalog or diagnosing its private proxy route.
claude-openaiUse when running Claude Code with an OpenAI subscription.
code-rag-setupInstalls or checks the dependencies of the code-rag skill (semantic code search with Qwen3-Embedding-0.6B + BM25): a pinned Python 3.12 venv made with uv under ~/.cache/code-rag and the model weights in the Hugging Face cache. Use when the user wants to set up, install, repair or check code-rag, or
code-ragSemantic search over a project's source code by meaning: finds the functions or files that implement a described behaviour when the exact names are unknown (\"where is the price of a cart item computed\", \"which code handles token refresh\"). Hybrid index: Qwen3-Embedding-0.6B + BM25 fused with RRF
compact-handoffMake a compact handoff of the CURRENT session, or of a NAMED OpenCode session (\"handoff of the OpenCode session X\"), so a new session can continue from it: one command writes HANDOFF.md and blobs/ with the full text of the long tool calls. Use whenever the user asks to write, prepare, make, update
compact-sessionCompact the active conversation in place, keeping recoverable history in local files. Use only when the user explicitly requests /apana:compact-session; unlike compact-handoff, this continues in the same session.
explica-clarA user-switched mode that makes every explanation in the session clear: summary first, few chunks, one real example, an analogy with a stated mapping, one flat diagram, signalled key points, wording matched to the reader, and a short understanding check. Use ONLY when the user explicitly asks for it
improve-skillImprove the user's own skills from how they were really used: digest the Claude Code transcripts into per-skill usage records (skill, version, trigger, problem signals), merge them across laptops, and propose evidence-backed changes to a SKILL.md or its description. Reporting is free: local files, a
install-claude-autoInstalls claude-auto, the wrapper that starts `claude`, writes a compact handoff when the context reaches 200k tokens and continues in a new session in the same terminal window, and puts it on PATH through ~/.zshrc. Use this whenever the user wants to install, set up, reinstall, update, repair or ch
jev-decision-loopA mode switched on by the user in which Jev (TypeSafe's judgment model, called over its direct API by a bundled script) takes EVERY decision of the coding session: before each tool call (every read and search too) you list 2-5 concrete next steps, Jev picks one, you do exactly that one. Needs TYPESA
logic-test-debugUse when writing or changing code, reading Boolean conditions, designing tests for combined conditions, or fixing bugs ("implement", "debug", "fix a bug", "scrie cod", "repară un bug", "depanează"). Do not use for non-code requests.
pr-convention-checkRead-only check of a diff against the live conventions of a repository: splits the diff into user-visible features and technical decisions, compares each decision element by element with the 3-4 most recent comparable commits on a reference branch the user names, and classifies every deviation as ju
review-diffReview a diff before it is committed: ask what to review when no argument is given (uncommitted changes, a commit, a pull request, something else), challenge every changed hunk, then every new class (why, strictly necessary, aligned with the project standard, in the right place, optimal), fix what i
review-planReview a plan before it is implemented: ask where the plan is when no argument is given, have two subagents run first (one finds the project's standards files, one looks up official best practices), challenge every step (why, strictly necessary, aligned with the project standard, assumptions true, i
scopeUse when the user asks to change code but wants strict scope: no extra files, tests only for behavior changed by this task, investigate resolvable repository questions before asking, and ask the user about intent rather than inventing requirements. Romanian: doar modificarea cerută, nu face nimic în
uninstall-claude-autoRemoves claude-auto (the wrapper that hands off to a fresh `claude` session at a token threshold) - its PATH block in ~/.zshrc and the ~/.claude/claude-auto/ folder - and leaves the rest of the rc untouched. Use this whenever the user wants to uninstall, remove, disable, undo or clean up claude-auto
uninstall-claude-copilotUse when the user asks to uninstall or remove claude-copilot, or preview its removal.
uninstall-claude-openaiUse when the user asks to uninstall or remove claude-openai, undo its installation, or remove its launcher or shell alias. Keeps OAuth credentials and shared dependencies. For installation or connection checks, use apana:claude-openai.
uninstall-claude-scopeUse when the user wants to uninstall, remove, disable, undo or clean up claude-scope, remove its zsh PATH entry, or restore after an installation. Romanian: dezinstalează claude-scope, scoate claude-scope din zshrc/PATH.
update-sessionsUpdates the apana plugin to the newest marketplace version and reloads it in every open Claude Code session on this Mac: runs `claude plugin marketplace update` and `claude plugin update`, finds the idle sessions in iTerm2, types `/reload-plugins` at their empty prompts and confirms each `Reloaded:`
variant-checkBefore writing or changing code that matches, filters, splits, labels or rewrites lines of text (a regex, a grep pattern, a parser, a classifier, a marker or label matcher over logs, transcripts, JSONL or CLI output), list every distinct variant of the matched token that really occurs in the data, h
/apana:explica-optiuniReface contextul și explică simplu rezultatul, necunoscutele și variantele de continuare. Numai la cererea utilizatorului.
node "${CLAUDE_PLUGIN_ROOT}/scripts/apana-hook.mjs" UserPromptSubmit
node "${CLAUDE_PLUGIN_ROOT}/scripts/apana-hook.mjs" PreToolUse
Talk like caveman. Cut filler. Keep technical details explicit.
22 skills · 1 command · 3 agents · hooks
Persistent memory for Claude Code: the next session starts where the last one left off.
29 skills · 1 MCP server · hooks
tasteskill 2.5.1: professional company websites at the figma.com and x.ai level, with one section per screen, real photographs, real product UI, effects kept for tech and gaming pages, and a real-browser checker and finish gate.
1 skill
Open-source AI job search that runs locally in your AI coding CLI: scan job portals, evaluate listings into a structured A-H report with a 1-5 score, tailor your CV, track applications
5 skills · hooks
Cross-session memory and token savings for coding agents.
6 skills · 1 agent · 1 MCP server · hooks
Package any website or local web build into a lightweight desktop app with Pake CLI.
1 skill
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.