Three hosts were watched loading it. The rest are documented paths.

Read this before reading the matrix

SecHelix separates format compatibility from model compatibility. The canonical workflow is a plain SKILL.md plus Markdown, JSON, and Python resources following the open Agent Skills format. A model does not need a SecHelix-specific API; it needs a coding-agent host that can load the skill or the equivalent files. So the question is never “is my model supported” — it is “does my host have a loader, and has anyone checked?”

Ground truth for every row on this page is docs/reference/compatibility.md. If this page and that file ever disagree, the file is right and this page is a bug.

VERIFIED — run here, recorded here

Three integrations were executed on a recorded date and machine. These are the only rows on this page backed by an observation rather than by documentation.

DOCUMENTED — the path is documented, the load was not observed

MODEL_COMPATIBLE — portable files, unverified loader

These rows claim the weakest thing the project can claim: the bundle uses only open-format frontmatter fields, so the files are usable. Whether the host discovers them is your check to run, not a claim being made here.

NOT_SHIPPED — deliberately absent

How the verified rows were tested

Executed on 2026-09-01, Windows 11 (10.0.26200), Python 3.14, Node 22.23.2, Claude Code 2.1.240, and the skills CLI via npx skills@latest. The bundle copy had every __pycache__ removed so no stale bytecode could mask a missing module, and the working directory was set inside the copy.

# Portable bundle, cold copy — all five entry points exited 0 from inside the copy
python scripts/validate_contract.py report examples/report.example.json
python scripts/security_gate.py examples/report.example.json --policy policies/default.json
python scripts/validate_gold_packs.py
python scripts/validate_knowledge.py
python -m reports.report_renderer examples/report.example.json --format markdown

# Claude Code plugin
claude plugin validate .
claude --plugin-dir . plugin details sechelix

# Agent Skills CLI, cold install into an empty directory
npx skills@latest add . --list
npx skills@latest add /path/to/SecHelix --skill sechelix

Nothing failed. The only ../ references in the bundle are relative Markdown links inside the Gold Pack README, which resolve within the bundle. The full transcript, including the exact output lines, is in docs/reference/compatibility.md.

Known packaging caveats

Two warts, recorded rather than hidden. Both affect what you actually get when you install, and neither is a security issue.

• The CLI installs the repository, not the portable bundle. npx skills@latest add … --skill sechelix matches the root SKILL.md and copies the whole repository tree — including tests/, evals/, artifacts/, and docs/ — into .agents/skills/sechelix/. The much smaller skills/sechelix/ bundle is what you get when you copy it yourself. Both work; only the second is minimal.
• The agents index file loads as an eighteenth “agent”. Claude Code scans the whole agents/ directory, so the README is loaded alongside the 17 real role profiles and the validator warns that it has no frontmatter. The documented manifest field does not fix it: on the tested Claude Code version, pointing it at individual files loaded zero agents, and pointing it at the directory made the plugin fail to load entirely. The field was left at its default and the wart is documented instead of papered over.

Why there is no marketplace.json

Claude Code does document marketplace.json, and a local marketplace is installable — so the mechanism is real. SecHelix still does not ship one, for three reasons that were measured rather than assumed.

• Adding marketplace.json beside plugin.json shadows plugin validation: with both files present, the validator checked only the marketplace manifest and stopped checking the plugin, its skill, and its agents. Losing that signal is worse than not having a marketplace.
• A marketplace whose single entry is the repository itself duplicates two install paths that already work, creating a second identity for one skill.
• Anthropic’s own documentation treats a combined plugin-and-marketplace repository as conflating two distinct roles rather than as a supported pattern.

If SecHelix later distributes more than one plugin, a marketplace belongs in a separate repository where it does not shadow this one's validation.

If your host has no skill loader

The portable fallback preserves one SecHelix workflow rather than maintaining incompatible prompts for every model vendor:

• Place skills/sechelix/ in the repository.
• Point the agent at skills/sechelix/SKILL.md.
• Use AGENTS.md for repository-level instructions.
• Keep scanner, browser, and database tooling outside the model-specific methodology.

The single recommended install command remains: npx skills@latest add omarmohelal/SecHelix --skill sechelix.