This is first-party agent skill setup that runs on ordinary CLI use, plus authenticated writes of remote guide text into project agent instruction files. It is not an npm lifecycle hijack, so it warrants a warning rather than a publish block.
Static reason
One or more suspicious static signals were detected.
Trigger
User runs any remits-cli command other than agent heartbeat-daemon; guide file writes additionally require components, test, token, tools, tool, or verify inside an authenticated account checkout.
Impact
Local coding agents pick up Remits skill and instruction files without a separate install --skills step. The guide content is controlled by the authenticated Remits host, so a compromised or malicious platform response can change agent instructions in that checkout.
Mechanism
main unconditionally calls syncInstalledSkills, which writes SKILL.md under the three agent home skill paths. ensureGuidesSynced posts the session token to /cli/guides and writes the instruction payload into CLAUDE.md, AGENTS.md, and GEMINI.md.
Evidence
index.jspackage.json~/.codex/skills/remits-cli/SKILL.md~/.claude/skills/remits-cli/SKILL.md~/.gemini/skills/remits-cli.mdCLAUDE.mdAGENTS.mdGEMINI.md