Agent Skills Toolbelt

My opinionated daily-driver skills for software engineering with OpenCode, Codex, Claude Code, and other agents that support the Agent Skills standard.

This is the small set I want available for every project: clear communication, better planning, root-cause debugging, parallel problem-solving, safer reviews, and an audit trail when work runs unattended. I recommend using it as a strong starting point, then adapting the skills to match how you build software.

Skills

Skill	Use it for
unslop	Remove generic AI phrasing and make writing direct and readable.
bro	Restate the previous response in plain human language.
grill-with-docs	Stress-test an idea while refining project terminology and decisions.
arena	Produce parallel solutions, judge them, and synthesize the strongest result.
diagnosing-bugs	Reproduce failures and find the root cause before applying a fix.
wizard	Generate a guided script for steps only a human can perform.
blast-radius	Find and verify what a local change could break elsewhere.
show-me-your-work	Keep a reviewable TSV decision trail for long-running work.

grill-with-docs depends on grilling and domain-modeling. Installing the whole toolbelt includes both. When selecting skills manually, select all three.

Install

The skills CLI supports OpenCode, Codex, Claude Code, Cursor, and many other agents.

Install the complete toolbelt. This is the recommended path and includes all eight daily drivers plus the two dependencies required by grill-with-docs:

npx skills add maxcodefaster/agent-skills-toolbelt --skill '*'

Or clone it and install from the local checkout:

git clone https://github.com/maxcodefaster/agent-skills-toolbelt.git
cd agent-skills-toolbelt
npx skills add . --skill '*'

Install every skill globally for OpenCode:

npx skills add maxcodefaster/agent-skills-toolbelt \
  --global --agent opencode --skill '*' --yes

Install the complete toolbelt for several agents:

npx skills add maxcodefaster/agent-skills-toolbelt \
  --agent opencode --agent codex --agent claude-code \
  --skill '*'

The skills CLI chooses the supported skill directory for each selected agent. Restart OpenCode after installation so a running session sees the new definitions.

Invoke

Agent Skills do not define one universal command syntax. Invoke the same skill using the mechanism provided by your agent:

Agent	Example
OpenCode	Use the grill-with-docs skill for this plan.
Codex	$grill-with-docs
Claude Code	/grill-with-docs

OpenCode discovers the installed definitions and loads them through its native skill tool. A slash command in OpenCode is a separate, OpenCode-specific configuration feature, so this harness-agnostic repository does not create slash-command wrappers.

Note

OpenCode loads skills on demand. To apply unslop to every response, add "~/.agents/skills/unslop/SKILL.md" to the instructions array in ~/.config/opencode/opencode.json, then restart OpenCode.

Repository layout

Each directory under skills/ is a standalone package:

skills/
  diagnosing-bugs/
    SKILL.md
    scripts/

Supporting files use paths relative to their skill directory. The skill bodies stay as close as possible to their upstream versions. Only Cursor-specific paths, model names, transcript locations, and delegation syntax were replaced with equivalent capabilities supplied by the current agent harness, including simple sequential fallbacks when a host cannot delegate work.

Sources

This collection carries selected skills from backnotprop/pstack and mattpocock/skills. Both sources are MIT licensed. The only changes make harness-specific instructions portable. See NOTICE.md for per-skill attribution.

Each skill carries its own LICENSE because installers copy skill directories independently and do not include this repository's root files. This keeps the required upstream MIT notice with every installed copy.

License

MIT