> ## Documentation Index
> Fetch the complete documentation index at: https://docs.erstan.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Partner Submission Checklist

> Package and review requirements for partner skills and templates.

Use this checklist before sending a skill or template for Erstan review. Upload validation currently enforces the E5 rule ids listed below; catalog-specific checks are marked manual until catalog lint lands.

## Package Shape

* Folder name matches frontmatter `name`.
* Frontmatter `name` is lowercase kebab case, 1 to 64 characters.
* Do not add `slug`; Agent Skills uses `name` as the package identity.
* `SKILL.md` is the entry file.
* Keep `SKILL.md` at 500 lines or fewer; move detailed material into `references/`.
* Include `license` and `compatibility` frontmatter when the package has licensing terms or platform constraints.
* Keep top-level frontmatter to `name`, `description`, `license`, `compatibility`, `metadata`, and `allowed-tools`.

Machine-enforced rule ids: `name_format`, `name_folder_mismatch`, `skill_md_too_long`.

## Description

* A description is required and may contain 1 to 1,024 characters.
* For better discovery, prefer 20 to 500 characters.
* Start with the concrete action, such as "Analyze", "Generate", or "Review".
* Include a trigger clause, such as "when the user..." or "use when...".
* Avoid generated boilerplate such as "Helps with...".

Machine-enforced rule ids: `description_length`, `description_boilerplate`, `description_missing_what`, `description_missing_when`. Only the Agent Skills 1-to-1,024 character requirement is blocking; the style and discovery checks are warnings during workspace import.

## Actions

* Declare every action with a stable lowercase key.
* Set `sideEffects: "none"` for read-only or deterministic local actions.
* Set `sideEffects: "external"` for actions that call outside services or mutate data.
* Declare required tools separately from prose instructions.
* Any write-capable workflow must declare that write actions require approval.
* Instructions must not direct the agent to execute writes inside a read batch or to page a full dataset through the conversation; bulk reads use the platform's read-only batch contract and each write is individually dispatched and approved.
* Describe required outcomes and invariants (counts, control totals) the agent must verify in code, rather than prescribing manual accumulation.

Machine-enforced rule id: `action_side_effects_invalid`. Write-approval intent is manually reviewed until catalog lint and runtime clone checks are fully enforced.

## Documentation Standard

Each submission must include:

* Who it is for.
* What it does.
* How it works.
* Setup requirements, including required integrations.

## Publisher And Review

* Include publisher attribution for partner or community content.
* State whether the package derives from any third-party source.
* Submit the zip package, then Erstan runs upload validation and catalog review.
* Listing requires a green validation result plus manual review for attribution, licensing, write behavior, and integration requirements.