Features · Spec 018

Multi-Agent Skill Integration

Two things bundled into one feature: extending Skill installation beyond Claude Code to five more coding agents, and making four of the canonical Skills actually consume the Context Engine's own output.

Six supported agents, one Skill format

Research (verified against each agent's own current documentation, not assumed) found that every one of these agents already scans the same open "Agent Skills" convention — a directory per Skill containing a SKILL.md with name/description frontmatter — the exact shape misterspec's own canonical Skills already use. Only the target directory differs:

No content transformation is needed for any of the five new adapters — each one is a byte-for-byte copy of the same install logic, differing only in which directory it targets. Two agents (Antigravity and Codex CLI) happen to scan the identical directory — installing for both in the same project is safe; neither corrupts the other's own files.

Example — install for a new agent:

misterspec init --agent cursor-agent --dir ./my-project
# Skills now live at ./my-project/.cursor/skills/<name>/SKILL.md

Four Skills now request a Context Pack first

/mister-implement, /mister-plan, /mister-tasks, and /mister-analyze each gained one new early step: call internal context with the matching intent before any other exploration.

Crucially, the Context Pack is a bootstrap optimization, never a sandbox: every one of these four Skills explicitly states the agent remains free to read further repository files whenever the pack alone isn't enough, and falls back to its own prior exploration approach if the internal context request fails for any reason.