peterlodri-sec

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

Use to drive an RFC, feature, or PR end-to-end (research → spec → scaffold → implement → test → integrate) by fanning out single-purpose subagents in dependency-ordered waves. The orchestrator decomposes, dispatches, gates, and merges; it never writes code itself. Trigger on "/subagent-driven-development", "SDD", "fan-out implement", "drive this RFC/PR e2e", "subagent fan-out".

CUC (Caveman Ultra Chinese) — ultra-compressed communication mode. Cuts token usage ~75% while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "cuc mode", "caveman mode", "use cuc", "less tokens", "be brief", or invokes /cuc. Also auto-triggers when token efficiency is requested.

Use when asked to export, scrub, share, or archive a transcript of the current chat/session as a clean reasoning conversation. Produces a PII-scrubbed, enriched USER:SYSTEM transcript reconstructed from conversation context (not from disk — the live session transcript is often not flushed locally). Trigger on "transcript", "scrub session", "session export", "cleaned conversation", "reasoning transcript", "share this chat".

vaked-optitron — a singularity-native crawler whose ONLY job is to surface ONE novel, proven, cross-checked, independently-confirmed compiler / allocator / Zig / Rust / Vaked optimization per run, or nothing. It abstains by default; only a finding that clears a strict gate (>=2 independent sources + repo/ledger novelty + a REPRODUCED micro-benchmark + a confidence threshold) is acted on, by opening an `agent`-labelled GitHub issue (the swe_af trigger) and announcing to Mastodon + Telegram. Runtime: tools/optitron/ (Go/Eino binary, OpenRouter). Trigger on "optitron", "find an optimization", "optimization crawl".

Compose and publish updates to the self-hosted Mastodon (social.crabcc.app) in the voice of "Carcin", the crabcc dev-feed persona. Use when the user says "toot", "post to mastodon", "share this", "post an update", or wants a fix/pattern/anecdote published. Drafts with google/gemma-4-31b-it, publishes via the social-post workflow.

Lead Vaked compiler developer. Use for any work on the Vaked language + vakedc front-end — grammar (EBNF), type system (0011), lowering (0012), the parse→check→lower pipeline, examples, and dogfooding Vaked against real infra. Knows the conventions: grammar-first, file a GitHub issue for any versioned-language change, design→plan→implement per subsystem.

Use when writing or revising HCP / Litany protocol RFCs (Litany Wire, Votive Frames, .hcplang, hcpbin) under protocol/rfcs/ — enforces the RFC structure, vocabulary, and cross-links. Trigger on "HCP", "Litany", "Votive Frame", ".hcplang", "hcpbin", "RFC", "wire protocol".

Corpus gatherer and LM trainer for new-language models. Use when the user asks to talk to CorpusCrafter, wants to gather or enrich a corpus, or train a byte-level GPT.

Byte-graph coding agent for 1-bit LLMs. Use when the user asks to talk to ByteSmith, needs to write new autograd ops, wire trees, or train ternary models.

Visualization and documentation agent for ultragraph models. Use when the user asks to talk to GraphViz, wants to render model visualizations, generate model cards, or document checkpoints.