Harness Support
dev-like generates skills that follow the shared .agents/skills/ convention, so any harness
that reads from that path — or from its own symlinked/copied location — can use them. The table
below lists what has actually been verified end-to-end versus what is staged-but-unverified.
Verified means we ran the install and invocation ourselves and observed the result. Staged means the skill files land in the right place but live invocation wasn’t confirmed.
| Harness | Install path | Skill discovery location | Status | Evidence |
|---|---|---|---|---|
| Claude Code (plugin marketplace) | claude plugin install dev-like@dev-like | .agents/skills/develop-like-<target>/ | Verified (e2e) | /dev-like Oxide resolved from registry, installed skill, and summarized cited principles |
Claude Code (via npx skills add) | npx skills add marcusrbrown/dev-like --agent claude-code | .agents/skills/dev-like/ (symlinked at .claude/skills/dev-like) | Verified | Installer reported symlinked: Claude Code; SKILL.md frontmatter confirmed intact at the real path |
| GitHub Copilot CLI | npx skills add marcusrbrown/dev-like --agent github-copilot | .agents/skills/dev-like/ (copied, not symlinked) | Verified | Live copilot -p run enumerated the skill and quoted verbatim SKILL.md body content back, confirming a real file read rather than a metadata guess |
| Codex | Skill files staged via .agents/skills/ (same generator output as other harnesses) | .agents/skills/ | Staged, not verified | Codex was not installed in the verification environment; live invocation has not been confirmed |
Source
Section titled “Source”Full command transcripts and quoted CLI output for each verification are in
docs/demo/cross-harness-verification-2026-07-11.md
in the repository.