Prompting Plugins

Commands

3clyp50

10

YAML-configured slash commands with text templates or Python hooks, plus an included skill that can create or update command files for you.

A0 Self-Improvement

TerminallyLazy

2

A0 Self-Improvement 2.3.1 adds a native Agent Zero interface with light/dark themes, outcome-aware learning, and failure-prioritized optimization, with governed replay, canaries, promotion, monitoring, and rollback.

Headroom Context Compression

emasoudy

2

Automatically compresses tool outputs in your agent conversations using Headroom AI. Saves 50-70% tokens on every agent turn by compressing file reads, command output, search results, and API responses. Works with any LLM provider. Zero configuration needed — install, run setup, done.

RLM

a0-community-plugins

2

Recursive long-context analysis for Agent Zero, with guided Docker Desktop setup, persistent workspaces, and sandbox diagnostics.

Runtime Cache

linkliti

2

Makes Agent Zero faster by remembering prompts and state in memory and refreshing them the moment anything changes on disk.

Chat Comments

agent-zero-plugins

1

Google-Docs-style comments for Agent Zero chats — select text in any message to attach a note, and one click bundles all comments into a single review request the agent addresses in one turn.

Compound Engineering

Jehu

1

34 compound-engineering skills - brainstorm, plan, debug, review and ship.

Guard System

PaoloC68

1

Security guards for tool execution and prompt safety. Blocks tools with flagged scan status and detects prompt injection patterns.

Misformat Guard

Olszalsik

1

Repairs misformatted chat-model responses (broken JSON tool requests) with a cheap utility-model call so Agent Zero never stalls or aborts on them. Defensive layers including a tool-repeat guard that breaks the reasoning death-loop where the agent re-emits the same failing tool call dozens of times. No core patch - pure extension hooks, with a live WebUI dashboard and per-agent configuration.

Caveman (compressed replies)

Olszalsik

0

Squeezes the assistant's own wording into short, accurate prose by having it answer in caveman-speak, across six intensity levels including Classical Chinese modes. Toggle per chat from the WebUI, with a slash command, or over HTTP. Reports the output it actually observed instead of an estimated saving, and ships a validated file compressor, 3 subagent profiles, 5 sub-skills, and 51 unit tests with 15 health-check self-tests.

SkillOpt Native

TerminallyLazy

0

Native skill training and offline learning for Agent Zero using Microsoft SkillOpt, with held-out validation, scoped proposals, reviewed adoption, and rollback.

Tiny Router

Nunezchef

0

Pre-classify every message in <10ms with a local DeBERTa-v3-small ONNX model (44M params). Routes low-value messages (thanks, ok, got it) away from the LLM — saving 30-50% on API costs at zero latency cost.