Skip to main content
Xerg keeps one canonical skill file. Codex, Cursor, Claude Code, Claude Desktop, Antigravity, and generic MCP clients use the same skill content; only their install or hosted MCP config differs.

​
Agent-guided setup

Give a terminal-capable agent this cold-fetch prompt:
The fetched file tells the agent to identify its host, ask before persistent installation and again before local analysis, run the skill’s reviewed CLI version’s JSON doctor, follow its version-checked shell-safe recommendation, and offer the first local audit without uploading automatically.

​
Install a reviewed revision

For Agent Skills hosts, first obtain and review a specific 40-character hexadecimal commit SHA from the public xergai/skills repository. Approve the installer download and skill installation, then replace the placeholder below with that reviewed SHA. Do not run the placeholder or substitute a branch, tag, or automatically selected latest revision. The pinned installer requires Node 22.20 or newer, independently of Xerg’s Node 22.13 minimum. An agent must not upgrade Node implicitly.
This fixes both the installer version and the skill’s source revision. It does not pin transitive dependencies or provide a cryptographic integrity guarantee. Agent selection and installation scope remain interactive; npx --yes only accepts the approved package download. Installation does not authorize local analysis, remote access, or hosted writes. If no reviewed SHA is available, stop the installer step or import an already-reviewed local skill folder through your host’s supported flow after permission. Do not automatically look up a replacement revision. Current-session use of the fetched skill still requires its separate permissions and must not persist it implicitly. Then ask your agent to audit your AI spend.

​
Canonical source

The source of truth is skills/xerg/SKILL.md, published to three places on every release: the public xergai/skills GitHub repo, the ClawHub listing, and the npm package bundle. The public mirror is:
The same bytes are available through both discovery conventions:
The v0.2.0 index includes a generated SHA-256 digest of the exact served bytes. The legacy index retains the name, description, and files shape used by Hermes.

​
Bundled package path

The published @xerg/cli package also includes a portable copy for agent tools that import skills from disk. Inside the installed npm package, the bundled copy lives at:
For a local project install, that usually resolves to:
For a global install, the same file lives inside the global npm package directory instead. Review the bundled skill independently before importing it: the skill’s reviewed CLI version can intentionally lag the skill release. Installing that older CLI does not provide the corrected current skill instructions.

​
What this file is for

The skill describes how an agent should use Xerg well:
  • when to reach for doctor --json, audit, explain, --compare, and --push
  • how to choose local, SSH, Railway, or multi-source audits
  • how to distinguish three evidence-strict monetary findings from seven neutral signals
  • how to explain every monetary finding before recommending remediation and treat repeated-chain diversity as neutral evidence
  • what to verify before finishing an audit task

​
Registry metadata

The skill frontmatter declares the runtime surface that ClawHub and agent registries need to understand:
  • xerg or npx as the CLI entry point
  • @xerg/cli as the npm package source
  • optional Xerg Cloud credentials and config paths for explicit hosted commands
  • optional SSH, rsync, and Railway dependencies for remote OpenClaw audits
These declarations are intentionally broader than the default local audit path because the same skill also documents opt-in hosted sync and remote audit workflows.

​
Important caveat

Installing @xerg/cli does not automatically register the bundled skill with every agent product. The reviewed-revision procedure above handles registration for supported agents; otherwise copy or link an already-reviewed skill folder into that product’s own skill directory or use its import flow after permission. Use agent clients for client-specific setup notes.