elohim-package-authoring
Manage Elohim-native agent skills by editing package metadata, treating other files as generated projections.
Install
mkdir -p .claude/skills/elohim-package-authoring && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/18579" && unzip -o skill.zip -d .claude/skills/elohim-package-authoring && rm skill.zipInstalls to .claude/skills/elohim-package-authoring
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
Author and maintain Elohim-native skills and agents from .epr-meta/elohim/packages, treating Claude and Codex files as generated projections.Key capabilities
- →Author or edit package JSON files under `.epr-meta/elohim/packages`
- →Treat `.claude/*` and `.codex/*` files as runtime projections
- →Regenerate projection fixtures with the package projection CLI
- →Generate runtime surfaces when local Claude/Codex files need refreshing
- →Verify projection drift before committing changes
- →Set `governance.eprRef` and other governance blocks for native packages
How it works
The skill manages Elohim-native skills and agents by treating package JSON files as the canonical source and Claude/Codex files as generated projections. It outlines a workflow for authoring, regenerating projections, and verifying fidelity.
Inputs & outputs
When to use elohim-package-authoring
- →Create new Elohim skill
- →Edit agent package definition
- →Sync runtime projections
- →Verify package fidelity
About this skill
Elohim Package Authoring
Use .epr-meta/elohim/packages as the canonical authoring surface for Elohim-native skills and agents. Claude (.claude), Codex (.codex), and Antigravity (.agents) are runtime consumers of those packages, never competing sources of truth.
Rules
- Edit package JSON first.
- Treat declared runtime files as generated projections for package-master and native packages. For an un-planted Claude-sourced package,
.claude/*remains the authored source while.codex/*and.agents/*are generated mirrors. - Treat every file below a skill root as part of the capability: relative
references/,scripts/,resources/, and other assets must live in the packageassetscollection, withmetadata.assetRefsmatching their paths exactly. Generated target directories are exact asset sets: removed package assets are pruned, and undeclared extras fail verification. - Run the package projection check after changing packages.
- Do not require a Claude source file for packages whose
metadata.sourceRuntimeiselohim-agent.
Workflow
- Author or edit the package under
.epr-meta/elohim/packages. - Regenerate projection fixtures with the package projection CLI.
- Generate runtime surfaces only when the repo intentionally wants local Claude/Codex/Antigravity files refreshed; use
--onlyfor scoped authoring. - Verify every projected
SKILL.mdand packaged asset before trusting the runtime surface. - Verify projection drift before committing.
Directionality contract
- Import (Claude→package) is certified byte-lossless. The standing fidelity gate (
verifySourceFidelityinpackage-projections.mjs, run byelohim-agent:packages:verify) assertsproject(import(source)) === source, byte-for-byte, for every Claude-sourced skill and agent. Everything below stands on that floor. - For an un-planted Claude-sourced package,
.claude/*is the authored source;.codex/*and.agents/*are generated mirrors. When such a package (metadata.sourceRuntime: "claude"with nometadata.master) has not been planted, the runtime files under.claude/are what a human or Claude edits directly. The package under.epr-meta/elohim/packages/is a certified mirror of that source AND the governance home — it carries anEprRefgovernance backref (metadata.governance.eprRef) that the runtime source file does not. Planting (next bullet) flips this. - The package-master flip is available, per package (2026-07-10). The
plant-eprfs-*skill family (plant-eprfs-skill,-agent,-hook,-agentdoc) plants a runtime-authored artifact: the package keepsmetadata.sourceRuntime(origin preserved) and gainsmetadata.master: "package"(authority rooted in the package). Once planted, the packageinstructions.bodyis the source of truth and.claude/.codex/.agentsare generated-and-clobbered projections — editing them directly is drift. The fidelity gate makes planting lossless: becauseproject(import(source)) === sourcebyte-for-byte, planting changes authority, not content. Both disciplines coexist in one tree: edit the package JSON for a planted (master: "package") or native package; edit.claudefor an un-planted Claude-sourced package. Adoption is one-at-a-time and operator-gated — plant one, prove it green, then the next. - Native packages are package-first. A package with
metadata.sourceRuntime: "elohim-agent"(this skill included) has no Claude source file — its package JSONinstructions.bodyIS the source of truth, and.claude/.codex/.agentsprojections are generated from it. Its governance block (metadata.governance) must be set by hand in the package JSON: thegovernanceFor()helper inpackage-projections.mjsonly threads through the Claude-import path (skillPackageFromClaude/agentPackageFromClaude) and never runs for native packages. Future native-package authors: write thegovernance.eprRef/policy/gates/ledgerblock yourself — nothing generates it for you. - Projection completeness includes assets. A green entrypoint with missing relative references is a false-green capability. Skill assets are package-owned bytes (
contentBase64), projected beside every runtime'sSKILL.md, verified byte-for-byte as an exact set, and recorded as their own eprfs derivation edges. - Runtime support follows a real ABI. A shared filesystem root does not prove feature parity. Antigravity skills use its documented
.agents/skills/<name>/SKILL.mdABI; subagents, commands, hooks, and MCP profiles require their own documented adapters beforeruntimeTargetsmay claim them. - The governance
EprRefis the offline floor anchor. It proves local fidelity and existence, gated byepr-meta-resolver+elohim-agent:packages:verify. Its ceiling — deep trust and REA value-flow reconciliation across peers — resolves through the p2p substrate, not this local tooling.
When not to use it
- →When editing `.claude`/`.codex` files directly for planted packages
- →When a Claude source file is required for packages with `metadata.sourceRuntime` as `elohim-agent`
Limitations
- →Does not require a Claude source file for packages whose `metadata.sourceRuntime` is `elohim-agent`
- →Editing `.claude`/`.codex` directly for planted packages is considered drift
- →The `governanceFor()` helper only threads through the Claude-import path and does not generate governance blocks for native packages
How it compares
This skill establishes a specific authoring surface and projection workflow for Elohim-native packages, where package JSON is the source of truth, which is distinct from direct editing of runtime projection files.
Compared to similar skills
elohim-package-authoring side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| elohim-package-authoring (this skill) | 0 | 2mo | No flags | Advanced |
| skill-creator | 128 | 4mo | Review | Advanced |
| skill-development | 17 | 9mo | Review | Intermediate |
| agent-identifier | 15 | 9mo | Review | Intermediate |
Try saying
Example prompts that trigger this skill in your AI assistant.
More by ethosengine
View all by ethosengine →You might also like
skill-creator
anthropics
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
skill-development
anthropics
This skill should be used when the user wants to "create a skill", "add a skill to plugin", "write a new skill", "improve skill description", "organize skill content", or needs guidance on skill structure, progressive disclosure, or skill development best practices for Claude Code plugins.
agent-identifier
anthropics
This skill should be used when the user asks to "create an agent", "add an agent", "write a subagent", "agent frontmatter", "when to use description", "agent examples", "agent tools", "agent colors", "autonomous agent", or needs guidance on agent structure, system prompts, triggering conditions, or agent development best practices for Claude Code plugins.
llama-factory
zechenzhangAGI
Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support
dify-dsl-generator
wwwzhouhui
专业的 Dify 工作流 DSL/YML 文件生成器,根据用户业务需求自动生成完整的 Dify 工作流配置文件,支持各种节点类型和复杂工作流逻辑
character-generator
Dexploarer
Generate complete elizaOS character configurations with personality, knowledge, and plugin setup. Triggers when user asks to "create character", "generate agent config", or "build elizaOS character"