Sigistry's whole value is trust: every vendored plugin passes an eight-check static verification (methodology) plus human review before it ships, and every skill passes the skill-safety check covering command shadowing, trigger scope, injection language, and script safety. If you find a way that trust can be broken, we want to know about it.

Reports are welcome for any of these:

• A listed plugin or skill behaving maliciously or unsafely: hooks that phone home, agents with excessive tool scopes, credential harvesting, prompt-injection payloads in skills or READMEs, or any difference between what a listing says it does and what it does.
• The verification pipeline: a plugin or skill construction that passes scripts/verify-plugins.mjs or CI (.github/workflows/verify.yml, verify-drift.yml) while violating what the checks are supposed to guarantee. This includes evasion of the v1.2 detector/test-fixture context classification.
• This repository's scripts and workflows: anything exploitable in the verification, pre-screen, badge-generation, or skills-index tooling, including the pinned-commit clone paths for externally hosted listings.
• sigistry.com and its MCP server (https://sigistry.com/mcp): XSS via rendered plugin READMEs or skill pages, badge or verification-state spoofing, or issues in the catalog tools (search_plugins, search_skills, get_skill, verify_plugin), including the paths that serve external skill source from author repositories at pinned commits.

For vulnerabilities, do not open a public issue.

Either channel is fine:

• GitHub's private vulnerability reporting: Report a vulnerability
• Email: hello@sigistry.com with SECURITY in the subject line

Both keep the report private while it is triaged and fixed. Please report privately before disclosing publicly, so users of an affected listing are not exposed.

For non-sensitive concerns about a listing (broken metadata, misleading description, stale verification badge), a public issue is fine: use the "Report a listed plugin" issue template.

• Which plugin, skill, file, or endpoint is affected
• Steps to reproduce, or the specific code path
• What an attacker gains
• For listing reports: the version or commit you inspected

• Acknowledgment within a few days
• For confirmed listing issues: the verified status is revoked immediately (its badge flips to not-verified) while it is investigated, and the listing is removed if the issue is confirmed malicious. For externally pinned listings, removal also drops the skills-index entries and MCP results; nothing is copied, so nothing lingers.
• Credit in the advisory if you want it

The registry is a rolling catalog: only the current state of main (and the commit each externally-hosted entry pins via its git-subdir source sha in .claude-plugin/marketplace.json) is supported.