Delegate bounded coding tasks to Kimi Code and review diffs

Lets an orchestrating agent hand a bounded coding task to the Kimi Code CLI (kimi) as a separate implementer, then review what it produced and land it itself. The defining constraint of this implementer is autonomy: in headless -p mode, Kimi always runs in auto permission mode and never asks for approval, and it outright rejects --prompt combined with --yolo, --auto, or --plan - so the relay passes none of them and offers no --read-only or --full-access option at all, unlike sibling delegate skills for other CLIs. There is no CLI-enforced read-only mode to fall back on; the only guarantee available is inspecting touchedFiles and the diff after every run, which the skill states plainly as "that diff, not a flag, is the guarantee of what changed." Dispatch wraps Kimi's headless prompt mode, captures its structured event stream, and writes a result.json to a temp directory by default so the reviewed repo stays clean; the relay never commits. The child process's working directory pins the workspace, and repeatable --add-dir flags add extra workspace directories beyond it. Rework resumes the most recent session (--resume-last) or a specific one (--session <id>) with just a delta brief rather than restating the whole task, since Kimi's session keeps its prior context.

Use it to delegate a bounded coding task to a separate kimi implementer and then review its diff yourself, when the user explicitly asked for delegation to this implementer. Do NOT use it when: the task is small enough to do inline and delegation overhead isn't worth it; the kimi CLI is not installed or authenticated; or the task genuinely needs a CLI-enforced read-only implementer, since headless Kimi has no read-only mode at all - use a different delegate skill instead when that guarantee matters.

The brief is the entire task as Kimi will see it - no chat history, no shared context beyond what it can inspect in the workspace - so it must state the goal, current state, what to change, what to leave untouched, the project's real gate commands, and a report contract telling Kimi not to commit. Dispatch:

node "<skill-dir>/scripts/relay.mjs" --brief brief.txt --cd /path/to/repo

Output is result.json with status (including kimi_unavailable when the binary is missing), finalMessage (Kimi's own report), and touchedFiles (the actual review starting point, since no permission flag constrains what Kimi did). A pre-run usage error exits 2 with no result file written; a missing binary exits 127 but still writes status: kimi_unavailable.

Wraps the Kimi Code CLI, installed via brew install kimi-code on macOS/Linux or the native installer from Moonshot AI's official docs, authenticated with kimi login (device-code flow) or /login in its TUI. Model selection uses default_model from Kimi's own config.toml unless --model <alias> names one of the human's already-configured aliases directly - the skill is explicit that an alias should never be invented, only reused from the user's existing config. Can apply a fleet lane's dials from a companion delegate-setup configuration.

Orchestrating agents that need to delegate a scoped coding task to Kimi Code and are prepared to rely on post-hoc diff review rather than any permission-mode flag, since none exists for this implementer in headless mode. Adapted from the community amElnagdy/delegate-skills repository under the MIT license; this catalog entry is a docs-only import, since the executable scripts/relay.mjs is not bundled here.