AbyssCNCommunity·12 Agent Skills Included

xihe

Verified multi-model coding pipelines with checkpointed graph execution

Runs coding tasks as verified graph pipelines where each step is checked by tests, exit codes, and a second model family instead of trusting self-reported completion. Covers code review, security audits, root-cause debugging, deep research, decision mapping, and video-to-notes extraction. Resumes interrupted runs from disk checkpoints so failed work never restarts from zero.

36|1

↗

npx skills add AbyssCN/xihe --all -g -y

Available:

Gives the agent strict working rules: verify every edit with typechecks and tests, block dangerous commands, ground every identifier in the real codebase, and distill one durable fact per session into memory.

All Skills in This Repository (12)

Pure Emerald Level Indicators

📦 In Repo

AbyssCN

commit

Automate Git commits with conventional messages and TypeScript checks.

council

Generate multiple design perspectives in parallel and select the best candidate.

verify

Run TypeScript compilation, tests, and builds for JavaScript projects.

handoff

Update the active plan and write a session log.

investigate

Guide systematic debugging through 8 phases to identify root causes.

start

Read next-steps file and git status to brief the current task.

recall

Retrieves past context from the memory layer for proactive decision-making.

skill-creator

Create and improve AI skills with SKILL.md frontmatter and subagents.

review

Conducts code reviews, security audits, and compliance checks on demand.

retro

Analyze git history to generate engineering retrospective reports.

caveman

Compress AI-generated text to reduce token usage by 75%.

dream

Consolidate an agent's raw events into durable memory facts.

Community

Intermediate

Frequently Asked Questions

FAQPage Schema

How to install xihe (oh-my-dag)?▼

Run `npx skills add AbyssCN/xihe --all -g -y` in your terminal to install all skills in this suite globally.

What problem does oh-my-dag solve?▼

It stops coding agents from claiming work is done without proof. Every task ends only when a real command like your test suite exits with the expected code, and a different model family double-checks subjective results.

Can interrupted runs be resumed?▼

Yes. Every finished node is saved to disk, and the resume skill reloads the plan, skips unchanged nodes, and only re-runs what actually failed.

Does it work with Claude Code and other coding agents?▼

Yes. It runs as an MCP server under your existing agent, and the skills follow the standard SKILL.md format usable in Claude Code and compatible clients.

Can I choose which model runs each step?▼

Yes. You can pin a specific model per pipeline node, mixing cheap models for high-volume steps and stronger models where mistakes are expensive.