Reliable agents aren't prompted.
They're built.
Agentiquette indexes, scores, and benchmarks the skills, instruction files, loops, and memory systems that make AI agents actually finish the job.
46 entries indexed · 45 fully scored · metrics refreshed 2026-07-07
Scored, not just listed
Every ranking is backed by a published rubric, timestamped data, and evidence notes. Adoption is capped at 5% of the score by design.
Repositories and frameworks
Verified live against the GitHub API, install instructions actually run, weaknesses stated alongside strengths.
A process-first skill framework for Claude Code: brainstorming, test-driven development, systematic debugging, and verification skills that…
Anthropic's first-party public skill collection: the de facto reference for SKILL.md structure, frontmatter conventions, and…
An opinionated 23-skill suite that staffs a solo developer's agent with the roles a startup would hire: CEO review, design review,…
140 domain skills plus 100+ database connectors that turn a coding agent into a working research assistant for biology, chemistry, medicine,…
A multi-harness plugin marketplace of role-specialized subagents (reviewer, architect, security, docs) installable across Claude Code, Codex…
Test harness for prompts, agents, and RAG pipelines: declarative test cases, assertions, matrix comparisons, and red-team scanning, runnable…
Below the repo: skills. Above it: patterns.
Individual skills get their own scorecards; the loops and memory architectures they implement get canonical definitions.
Blocks the agent from claiming work is complete, fixed, or passing until it has run verification commands and confirmed output. Evidence…
Forces root-cause investigation before any fix: reproduce, hypothesize one cause at a time, test the hypothesis, then patch. Bans…
Requires a failing test before implementation code for any feature or bugfix. The loop is red, green, refactor, with the agent forbidden…
The agent produces an explicit plan, executes one step at a time, and verifies each step's output against stated criteria before proceeding.…
Long-running work is segmented at durable checkpoints: after each segment the agent writes state (what is done, what is next, open…
Semantic memory as flat files: one fact per file with dates and metadata, plus a one-line-per-fact index loaded every session. Fact bodies…
From the Lab
One-shot prompting vs plan-execute-verify on bug-fix tasks
A plan-execute-verify loop will complete more of a fixed bug-fix task set than one-shot prompting with the same model and tools, at higher cost per task, with a lower false-success rate.
Status: pre-registered · protocol published 2026-07-07 · N=10 runs per task per condition
All benchmarksWhy pre-register?
Rubrics written before runs cannot be bent to fit results. Every Agentiquette benchmark publishes its method first, its results second, and its limitations always.
Templates
Copyable starting points with annotated fields. Never gated.
Maintain a skill repo, loop pattern, or instruction framework?
Submit it for indexing and scoring. Free, reviewed by humans, corrections welcome.
Submit for scoring