K
Star Keystone

Flexible Entry.Honest Exit.

Keystone is a workflow skill bundle for Claude, Codex, and Gemini. Enter from any stage, but only claim done with fresh evidence.

18 Skills
Honest Exit
Flexible Entry

The Biggest Risk in AI Agents

Not slow output, but false confidence. Most agents confidently output code regardless of whether tests pass.

Pretending to be done

Agent says done, but tests didn't run, edge cases weren't covered, verification wasn't actually executed. Verbal done, not evidence-backed done.

No evidence

The Keystone Way

Enter from any stage, but claim done only with fresh evidence. No evidence, no claim of done.

Flexible Entry
Honest Exit

Forced from scratch

Must start every task from step one? Real work starts mid-stream. Post-review, mid-bug-fix, incident response.

Forced from scratch

Why Keystone

Most AI agents are optimized for one thing: looking done. What Keystone adds is a different layer: making every done actually have fresh evidence.

Flexible Entry

Enter from any reasonable stage. Mid-implementation, post-review, incident response — no need to start from scratch every time.

Honest Exit

No fresh evidence, no claim of done. Tests passing, review finding real issues, canary stable — every done needs verifiable evidence.

Unified Decision Contract

Every skill outputs decision, confidence, rationale, next_skill. Cross-session, cross-agent readable state.

Shared State

Read/write `.keystone/state.json` across sessions. Multi-person collaboration, handoff recovery, cross-stage entry — everyone works from the same shared state.

Adapts to How You Work

Whether you're starting fresh, resuming mid-task, or just need a verification checkpoint — Keystone has an entry point.

Full Feature

Start from scratch: `/ks-orchestrate` → roundtable → writing-plan → test-first → implement → verify → review → ship → release

Hard Constraint
Standard Step
Click any skill for details

18 Skills, 5 Layers

All skills follow the same protocol: Decision Contract, Honest Exit, Shared State, Telemetry. Work seamlessly across Claude, Codex and Gemini. (Same skill names, prefixed with `ks-*` instead of `pf-*`)

Orchestration

orchestrate · handoff. Route to the right entry, manage session context, enable clean handoffs.

Decision

roundtable · brief · writing-plan · review. Brainstorm, compress context, create plans, enforce quality gates.

Execution

test-first · implement · verify · diagnose · bug-triage. Build with boundaries, verify with fresh evidence, find root causes.

Operations

qa · pr-prep · ship · release. Validate in production, prepare context, deploy safely, close explicitly.

Support

help · docs-writer · knowledge. Document outcomes, build and retrieve reusable patterns.

Install in 30 Seconds

One-liner installation, unified skill naming, immediate access across Claude, Codex, and Gemini.

bash
1# Copy this one-liner
$curl -fsSL https://raw.githubusercontent.com/heyallencao/Keystone/main/scripts/install.sh | bash
2# Paste into your terminal
3# Restart your AI host (Claude/Codex/Gemini)
# Start with any ks-* skill
Claude:Claude: /ks-help
Codex:Codex: use /ks-orchestrate
Gemini:Gemini: use /ks-brief

One-liner Setup

30 seconds, no complex configuration

Unified Naming

Same `/pf-*` prefix everywhere

Works Everywhere

Claude, Codex, Gemini seamlessly