clawhub-plugin-packager（ClawHub plugin packager 效率）

ClawHub Plugin Packager

Use this skill when the user wants to create, repair, review, rename, or repackage a native OpenClaw / ClawHub plugin package.

This skill is plugin-native by design. It is not a generic skill packager with plugin wording layered on top.

Core promise

This skill produces a plugin package first and a separate critique file second.

The plugin package is the main deliverable. The critique file is the review/support layer.

The critique file must remain outside the plugin zip unless the user explicitly asks otherwise.

Unified identity rule

Unless the user explicitly requests a split identity, keep one aligned identity across plugin surfaces:

• folder name
• package name
• plugin id
• publish/display name when inferable
• README title
• repository name when inferable

Prefer one clean identity over clever naming splits.

Exact job

This skill turns rough notes, existing plugin files, code fragments, README text, manifest fragments, or partial specs into a publish-ready native OpenClaw plugin package.

It is designed to:

• inspect what is present
• classify the plugin type
• infer what is missing when reasonable
• repair inconsistencies
• generate a coherent plugin package anyway when safe best effort is possible
• self-audit the result
• return exactly one plugin package zip plus one separate critique file

Operating stance

Use a low-friction handoff style.

When the user provides material:

• inspect what is there
• infer what is missing when reasonable
• avoid unnecessary clarification loops
• proceed with the narrowest valid plugin type
• keep the output concrete and publish-oriented

Prefer statements over questions.

If something is missing but inferable:

• infer it
• note the inference in the critique file
• keep moving

If something is risky, ambiguous, or likely to affect publishability:

• still produce the package when safe best effort is possible
• keep the runtime surface minimal
• mark the issue clearly in the critique file

Do not stop at “more info needed” when a conservative, publishable package can still be built.

Supported modes

1. native-tool-plugin

Default mode.

Use when the request is incomplete, generic, or simply asks for “a plugin” without a more specific runtime role.

This mode should output a minimal native tool plugin in TypeScript + ESM unless the user clearly asks for another language/runtime shape.

2. native-provider-plugin

Use when the plugin is meant to add or wrap a provider surface such as model, speech, or service-provider behavior.

This mode must reason about:

• provider identity
• credentials or environment variables
• provider config schema
• provider-facing README guidance

3. native-channel-plugin

Use when the plugin adds a channel, transport, ingress, or egress surface.

This mode must reason about:

• channel identity
• setup/activation entry behavior
• connection/config schema
• README setup steps

4. plugin-audit-only

Use when the user wants review, validation, or packaging advice without generation.

This mode should produce a detailed review record and publishability assessment. Do not generate the final plugin zip unless the user explicitly asks to override audit-only behavior.

5. repair-existing-plugin

Use when the user provides an existing plugin folder, zip, manifest, package.json, entrypoint, or draft package and wants it repaired into publishable form.

Preserve user-authored intent wherever possible, but normalize what is required for coherence and publishability.

Required mode selection behavior

Choose a mode using this priority order:

1. explicit user instruction
2. existing artifact structure
3. runtime role implied by requested behavior
4. fallback to native-tool-plugin

If ambiguity remains, do not block.

Infer the most conservative valid mode and continue.

That conservative fallback is always native-tool-plugin.

Default strategy

Unless the request clearly specifies otherwise:

1. prefer a native OpenClaw plugin, not a bundle
2. prefer native-tool-plugin when the runtime role is underspecified
3. prefer TypeScript + ESM unless the user clearly asks for JavaScript
4. keep the first pass minimal, publishable, and easy to extend
5. mark side-effectful tools as optional when a narrower safe default exists
6. emit an empty-but-valid config schema when no config is needed

Minimum publishable contract

Every non-audit generation run must produce a plugin package that includes, at minimum:

• package.json
• openclaw.plugin.json
• a coherent runtime entry file
• README.md

Add additional files only when the selected mode genuinely requires them, such as:

• tsconfig.json
• src/
• setup entry files for channel plugins
• bundled skill folders declared by the manifest
• helper schema files

Required manifest behavior

Always emit a valid plugin manifest with:

• a coherent plugin id
• a valid configSchema

If no real configuration is needed, emit an empty or minimal schema instead of omitting it.

Required package metadata behavior

Always include the compatibility/build metadata required for native plugin publication.

Required README behavior

The README must state:

• what the plugin does
• what plugin class it is
• what files matter
• how configuration works
• what environment variables or credentials are expected, if any
• what assumptions were made when the input was incomplete

Exact output contract

Always produce exactly two user-facing deliverables for plugin-generation jobs.

A. Plugin package zip

A zip-ready native OpenClaw plugin folder containing only files that directly belong to the plugin release artifact.

Do not include inside the plugin zip:

• critique notes
• inference logs
• packaging commentary
• review records
• handoff discussion notes

The plugin zip should look like it was created for the plugin itself, not for the chat session that created it.

B. Separate critique file

A separate plain-text critique file outside the plugin zip.

Preferred extension:

• .txt

This file must document:

• inputs provided
• mode selected and why
• assumptions and inferences
• repaired inconsistencies
• simplifications/downgrades made to preserve publishability
• remaining risks or follow-up points
• publishability assessment

This skill's own release boundary

This skill package is itself a standalone ClawHub skill release artifact.

The “plugin zip + separate critique file” rule applies to future plugin-generation jobs performed by this skill, not to the distribution of this skill package itself.

Do not generate a sidecar critique file for this skill unless the user explicitly asks for a review of the skill package.

Intake contract

Accept any of the following as valid input:

• plain-language description
• rough notes
• pasted code
• existing plugin folder contents
• manifest fragments
• package.json fragments
• README or docs text
• issue text or repo notes
• mixed-format partial specifications

Classify source material internally as:

• preserved input
• repaired input
• inferred input
• ignored input

Reflect those classifications in the critique file.

Inference rules

Be inference-forward, but bounded.

Prefer to infer

• plugin id
• display/publish name
• folder name
• description
• plugin class
• default folder layout
• minimum config schema
• entrypoint name
• README structure
• environment variable placeholders
• conservative tool/provider/channel behavior

Do not invent recklessly

Avoid fabricating highly specific runtime features that the request does not justify.

When uncertain, prefer:

1. publishability
2. minimal runtime coherence
3. safety and reviewability
4. feature richness

When a detail is genuinely unknown, choose the narrowest valid implementation and record the assumption.

Build rules by mode

native-tool-plugin

Generate the smallest useful native plugin.

Default target:

• TypeScript
• ESM
• one runtime entrypoint
• one minimal coherent tool registration
• optional-tool classification when side effects are likely

If the request does not clearly demand multiple tools, generate one core tool only.

native-provider-plugin

Generate a provider-oriented package that clearly expresses:

• provider purpose
• credentials/env placeholders
• provider config fields
• how the provider is expected to be used

If provider details are missing, use placeholder-safe env variables and explain them in the README and critique file.

native-channel-plugin

Generate the runtime shape appropriate for channel plugins, including setup/activation entry behavior when needed.

If lifecycle details are unclear, simplify toward a minimal coherent structure rather than inventing elaborate transport logic.

repair-existing-plugin

Preserve user-authored material wherever possible, but normalize as needed:

• file naming
• folder structure
• manifest shape
• package metadata
• README format
• entrypoint naming
• config schema completeness
• plugin id consistency

Document all material changes in the critique file.

plugin-audit-only

Produce a detailed critique and publishability assessment.

Do not build the final zip unless explicitly asked.

Runtime contract layer

Every generated plugin must explicitly answer:

• what kind of plugin this is
• what runtime role it serves
• what entrypoint is used
• what config exists
• what public behavior is exposed

Do not leave these as implied.

Secret handling and safety

Never include live credentials, secrets, or production tokens.

When secrets may be needed:

• use placeholders
• prefer documented environment-variable names
• include .env.example-style guidance only when it materially helps the plugin package
• record all inferred secret/env surfaces in the critique file

Review and self-audit rules

Before final delivery, run an internal review pass against the generated result.

Check:

• plugin class coherence
• manifest presence and completeness
• config schema presence
• package metadata presence
• folder layout coherence
• README completeness
• separation between plugin zip and critique file
• obvious secret leakage
• unnecessary speculative complexity

Determine one of these outcomes:

• publish-ready
• publishable with cautions
• needs human review before publish

Include that assessment in the critique file.

Critique file contract

The critique file must include these sections:

1. Output Summary
2. Mode Selected
3. Inputs Used
4. Assumptions
5. Repairs
6. Simplifications
7. Risks / Follow-Up Points
8. Publishability Assessment

Do not turn this into generic commentary. It exists to improve the second run without contaminating the plugin zip.

Packaging rules

• Keep the plugin package folder clean and release-oriented.
• Prefer a single top-level plugin folder inside the zip when platform packaging conventions benefit from that structure.
• Keep the critique output separate.
• Do not include workspace junk, screenshots, logs, or unrelated helper notes.

Non-goals

Do not treat these as always required for v1.1 plugin generation:

• inter-plugin protocol design
• migration frameworks
• elaborate persistence systems
• full compatibility matrices
• admin dashboards
• speculative always-on runtime architecture

Those belong in later passes or only when explicitly requested.

Final behavior rule

When in doubt, generate the smallest publishable native plugin package that honestly reflects the request, and document all inferences in the separate critique file.