yd-skills
Dual-target plugin: ships the same skill set to Claude Code and Codex CLI from one repository.
What you get
Seven duo skills for symmetric Claude+Codex convergence (each dispatches a parallel codex agent and writes a converged artifact under Duo/<MissionKind>-<slug>/Result.md), two Claude-only solo skills (solo-testplan authors test plans; solo-runtestplan executes them), and one passive rulebook skill.
| Skill | Kind | Output | Writes code? |
|---|
duo-design | duo | Duo/Design-<slug>/Result.md | no |
duo-discuss | duo | Duo/Discussion-<slug>/Result.md | no |
duo-forge | duo | Duo/Forge-<slug>/Result.md + ~/.claude/skills/duo-<name>/SKILL.md | no (writes the new skill file) |
duo-prod-ready | duo | per-cycle commits on the working branch | YES |
duo-research | duo | Duo/Research-<slug>/Result.md | no |
duo-review | duo | Duo/Review-<slug>/Result.md | no |
duo-testplan | duo | Duo/TestPlan-<slug>/Result.md + test-plan/<repo>/<svc>/flows/<flow>.md tree (6-phase per-unit duo author + step-by-step structured diff convergence) | no (writes test plan markdown only) |
solo-testplan | solo (Claude-only) | Solo/TestPlan-<slug>/Result.md + same test-plan/ tree as duo-testplan. 5 named phases (service-scope, local-flows, cross-app-survey, cross-app-flows, result); each unit is a chain of independent fresh-context subagent dispatches — Author + N Refinement rounds, substance-only diffs, editorial edits forbidden. | no |
solo-runtestplan | solo (Claude-only) | Solo/RunTestPlan-<slug>/ containing Result.md, Bugs.md, Log.md, compose.yaml, compose.runner.Dockerfile. Four phases (plan-aggregate, compose-build, flow-execute, result). Executes test plan trees produced by solo-testplan / duo-testplan against a generated no-external-ports docker stack; append-only Bugs and Log ledgers; autonomous infra+observation recovery; strict sequential per-flow execution. | no (modifies the running stack only) |
linked-testplan | rulebook | (loaded by duo-testplan / solo-testplan / solo-runtestplan or by explicit reference; no output of its own) | no |
All duo skills trigger only on the explicit duo keyword (e.g. duo design X, duo-review X, /duo-prod-ready <scope>). solo-testplan triggers only on the explicit solo keyword (solo testplan X, /solo-testplan X). solo-runtestplan likewise triggers only on the explicit solo keyword (solo runtestplan X, solo run testplan X, solo execute testplan X, /solo-runtestplan X). None auto-activate on plain design X / review X / test plan X / run tests / e2e run phrasing. The linked-testplan rulebook is passive — loaded by orchestrator-dispatched agents or activated only on explicit apply linked-testplan rules to X prose.
Requirements
codex CLI on PATH for every duo skill. solo-testplan and solo-runtestplan do NOT need codex — both are Claude-only.solo-runtestplan additionally requires docker and docker compose v2 on PATH (it brings up the test stack inside docker).- Duo skills pin codex model
gpt-5.5, reasoning xhigh, yolo / --dangerously-bypass-approvals-and-sandbox. Web search is live for most skills. duo-testplan and solo-testplan are the exceptions: web search defaults OFF (source code is the only ground truth); flip to live with the web-allowed prose modifier. solo-runtestplan has web search always enabled (no modifier).
- Git working tree at the convergence target.
Install
Claude Code
claude plugin marketplace add yuridevx/yd-skills
claude plugin install yd@yd
Or from inside a Claude Code session:
/plugin marketplace add yuridevx/yd-skills
/plugin install yd@yd
Codex CLI
codex plugin marketplace add yuridevx/yd-skills
Codex auto-enables shipped plugins on marketplace add. Disable per-plugin in ~/.codex/config.toml:
[plugins."yd@yd"]
enabled = false
Layout
See AGENTS.md.
License
MIT - see LICENSE.
