Skip to content

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.

HarnessInstall pathSkill discovery locationStatusEvidence
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)VerifiedInstaller reported symlinked: Claude Code; SKILL.md frontmatter confirmed intact at the real path
GitHub Copilot CLInpx skills add marcusrbrown/dev-like --agent github-copilot.agents/skills/dev-like/ (copied, not symlinked)VerifiedLive copilot -p run enumerated the skill and quoted verbatim SKILL.md body content back, confirming a real file read rather than a metadata guess
CodexSkill files staged via .agents/skills/ (same generator output as other harnesses).agents/skills/Staged, not verifiedCodex was not installed in the verification environment; live invocation has not been confirmed

Full command transcripts and quoted CLI output for each verification are in docs/demo/cross-harness-verification-2026-07-11.md in the repository.