/plugin-submission

name: plugin-submission
description: Govern discovery, eligibility, and submission of a plugin or skill to official directories, registries, marketplaces, and curated GitHub lists. Use when preparing or sending plugin listings, marketplace forms, directory PRs, or repository recommendations; require current policy evidence and a final human confirmation before every external submission.

Plugin Submission

Treat every public listing as an external representation of the project. Submit only an artifact that the destination's current policy accepts, and retain a receipt for every send.

Workflow

1. **Classify the destination and artifact.** Determine whether it accepts a full plugin, a repository, or one standalone `SKILL.md`. Do not describe a multi-skill plugin as one skill, or split a skill bundle without an explicit portable artifact. 2. **Read the current primary policy.** Use the destination's official submission instructions or `CONTRIBUTING.md`; record its URL, access requirements, licence and validation rules, and any limits on automated or AI-assisted submissions. 3. **Validate eligibility.** Check the public candidate URL, release/tag, licence as detected by the hosting platform, manifest/schema validity, documentation, and the destination-specific validator. Treat a mismatch or unmet requirement as `blocked`, not as an invitation to bypass it. 4. **Prepare one destination matrix.** For each target, record: destination, artifact, policy evidence, status (`ready`, `blocked`, `deferred`, or `human-only`), exact payload, and rollback action. Reuse the canonical repository description; minimise personal data and never add an email without explicit permission. 5. **Stage, then confirm.** Fill or draft public data only after the operator has approved the destination. Immediately before clicking a submit control, opening a PR, or sending a recommendation, restate the exact target and payload and obtain a final human confirmation. A single confirmation may cover multiple named sends only when the full set and payloads are shown together. 6. **Send through the permitted route.** Use a form, PR, issue, or API only when the policy permits that route. Never act where a policy requires a human-authored recommendation, prohibits AI-assisted submissions, or requires credentials/access the operator has not supplied. 7. **Verify and record the receipt.** Capture the submission URL/identifier and status. If a listing is rejected or stale, record the reason and next eligible action. Roll back a repository contribution by closing its PR/issue when appropriate; external directory records require the destination's own removal/update path.

Submission Gate

Do not send until all applicable checks pass:

• The target policy is current and supports the proposed artifact.
• The public repository and exact release are accessible.
• Project licence intent and platform-detected licence agree.
• Required validation has passed, including the destination's validator where available.
• The project has not overstated supported platforms, usage, security review, or

affiliation.

• The operator has given final confirmation for the exact outgoing entries.

Outputs

• A destination matrix with evidence URLs and eligibility verdicts.
• A staged payload for every ready target.
• Submission receipts, or an explicit blocked/deferred reason and owner.

Anti-rationalisation

| Temptation | Gate | |---|---| | “It is close enough to call the bundle one skill.” | Submit the artifact type the directory actually accepts, or create a deliberate portable export first. | | “The README says Apache, so platform metadata does not matter.” | Directories evaluate the public repository; resolve detected-licence mismatches before broad listing. | | “The user asked to submit, so the final click is implied.” | A final confirmation binds authority to the exact external payload. | | “A PR is harmless discovery.” | A public PR is a durable representation of the project and must follow the contributor policy. | | “We can automate a human-only form.” | Human-only and no-AI policies are hard stops. |