The Complete AI Development Toolkit for Claude Code. 106 skills, 36 agents, 171 hooks. Install `ork` for stable (v9.x), or `ork-alpha` for the v10 line, which ships daily.
This repository also configures its own agents. See what orchestkit tells them →
Cursor rule Cursor
Adversarial Refutation Engine. Shared blind-refuter protocol that removes self-preferential bias from evaluative skills — the producer of a finding can't be its own fair judge.
Cursor rule Cursor
Anti-Sycophancy Response Protocol. Prevents performative agreement that buries critical findings.
Cursor rule Cursor
Status: confirmed CC 2.1.81–2.1.143 · upstream bug filed anthropics/claude-code#60155 · prior incident orchestkit#1629.
Cursor rule Cursor
Chart Encoding Standard. Governs the DATA-MARK layer of every chart OrchestKit emits (playground, dashboard, infographic, MCP visual, multi-surface). ork does not own chart color science — it rents it from CC's /dataviz. The difference between an accessible, validated chart and an eyeballed rainb...
Cursor rule Cursor
Code Quality Thresholds. Keep functions under 50 lines.
Cursor rule Cursor
Docstring Standards. All public functions (no prefix) should have """docstring""" documentation.
Cursor rule Cursor
Evidence Replay Gate. Producer findings in evaluative pipelines are inadmissible until the orchestrator has deterministically replayed their cited evidence, so hallucinated citations never reach a verdict, score, or report.
Cursor rule Cursor
License Compliance. When installing new dependencies, check their license before adding to production projects.
Cursor rule Cursor
Playground Visual Standard. Governs the visual quality of every playground OrchestKit emits. The difference between a considered product demo and a generic AI dashboard is the specific values in this file.
Cursor rule Cursor
Shared Cross-Cutting Rules. Cross-cutting evidence requirements for all skill completion claims.
Cursor rule Cursor
Shell Count Correctness. Counting lines or files in shell is the most common thing our audit and telemetry scripts do, and every recipe below has already shipped a wrong number in this repo. A count that is silently low reads as a passing gate.
Cursor rule Cursor
Untrusted Input Quarantine. Prompt-injection defense for skills that read GitHub issue/PR/CI content — separate the agent that READS untrusted text from the agent that ACTS.
Cursor rule Cursor
Verification Gate. Prevents premature completion claims across all skills.
Cursor rule Cursor
Visual Style, glyph renders in chat. Reset 2026-09-17 by operator word after seeing a peer's disk-cleanup render.
Cursor rule Cursor
Long-Run Protocol. Long multi-agent runs finish at a named line, stop only where a human is needed, and never pass an unchecked claim as fact.
Cursor rule Cursor
Done Sign-off. Closes a task only when the person accepts the verified result, never on the agent's own word.
Cursor rule Claude CodeCursor
Target Resolution. A target-taking skill reads a pronoun back from the conversation instead of refusing it.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: