Disclaimer: grimTalk is not affiliated with Hasbro or Transformers.

grimTalk

Turse Bot Warrior communication mode for AI coding agents based on caveman. Cuts token usage ~65% while keeping technical accuracy.

Warrior prose. No fluff. Code strong.

What it does

• Grim mode — terse, fragment-heavy bot warrior responses.
• grim-commit — Conventional Commits messages, ≤50 char subject
• grim-review — one-line PR findings with severity emoji
• grim-compress — compresses .md memory/doc files to caveman-like prose (~46% input token savings)
• grim-stats — token usage + savings for current session
• Din-bots — subagent presets that return compressed tool results, shrinking main context per delegation

Install

Agent	Quick start	Full guide
Claude Code	bash agents/claude/install.sh	docs/install-claude.md
Kiro CLI	bash agents/kiro/install.sh	docs/install-kiro.md

All installers support --force (reinstall) and --uninstall. Restart your agent after install.

Usage

grimTalk is both a skill (slash command) and a custom agent. The /grimTalk slash command activates the skill. Sub-skill behaviors are triggered by natural language keywords — the model matches them from the injected SKILL.md.

Mode control

Trigger	What happens
/grimTalk	Classic grim — fragments, short synonyms

Technical terms, code symbols, function names, API names, error strings: never abbreviated at any level.

Natural language also works: "din-bot mode", "talk like grim", "less tokens", "smash word".

Deactivate: say stop grim, stop din-bot, or normal mode.

Sub-skills (keyword-triggered)

Say	What happens
"write a grim-commit" / "commit message"	Terse Conventional Commits message, ≤50 char subject
"grim-review this PR" / "code review"	One-line findings with severity emoji
"grim-compress file" / "grim-compress <file>"	Compress .md file to caveman prose (runs python3 -m scripts)
"grim-help" / "what grim commands"	Quick-reference card
"grim-stats"	Token usage + savings for current session
"use din-bots" / "delegate to subagent" / "save context"	Subagent delegation guide

Project structure

SKILL.md                  — skill definition loaded by agents
references/
  grim-commit.md          — commit message rules
  grim-review.md          — code review rules
  grim-compress.md        — compression rules
  grim-help.md            — quick-reference card
  grim-stats.md           — token stats rules
  grim-style.md           — style guide & rules for agent prompt
  din-bots.md             — subagent delegation guide
scripts/
  compress.py             — compression orchestrator
  detect.py               — file type detection
  validate.py             — output validation
  benchmark.py            — compression benchmarks
  cli.py                  — CLI entry point
agents/
  kiro/                   — Kiro CLI installer + hooks
  claude/                 — Claude Code installer + hooks
docs/
  install-claude.md       — Claude Code full install guide
  install-kiro.md         — Kiro CLI full install guide