# Kimi First

> Route implementation work-orders to Kimi k3 lanes via the Kimi Code CLI; Claude specs, orchestrates, reviews, verifies. Use when a task is a frozen-spec build, mechanical migration, test fill, bulk sweep, or CI fix — anything that reads as a work order rather than a design problem.

- **Type:** Skill
- **Install:** `agentstack add skill-boringmarketer-kimi-first-kimi-first`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [boringmarketer](https://agentstack.voostack.com/s/boringmarketer)
- **Installs:** 0
- **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [boringmarketer](https://github.com/boringmarketer)
- **Source:** https://github.com/boringmarketer/kimi-first/tree/main/skills/kimi-first
- **Website:** https://boringmarketing.com

## Install

```sh
agentstack add skill-boringmarketer-kimi-first-kimi-first
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Kimi First

Claude Code sessions only. Rationale: Kimi k3 lanes run flat-rate on the Kimi
membership and are strong at implementation throughput; Claude's edge is judgment,
specs, orchestration, review. Kimi types, Claude thinks and verifies. (Adapted from
[codex-first](https://github.com/steipete/agent-scripts/blob/main/skills/codex-first/SKILL.md)
in steipete/agent-scripts.)

## Route

**Delegate to Kimi** (default for hands-on volume):
- implementation from a frozen spec; refactors; mechanical migrations/sweeps
- bug fixes with a known repro; test writing; coverage fills
- CI fixes, dependency bumps, scripts/tooling
- bulk exploration where raw reading ≫ the answer

**Keep in Claude (never delegate):**
- design, architecture, naming, product judgment — tasks where writing the spec IS the work
- tiny edits ("$P" 
EOF
(cd  && kimi -p "$(cat "$P")" > "$SCRATCHPAD/kimi-last.md" 2>/dev/null)
```

- `-p` is headless: no approvals asked, tool calls run under auto permission; static
  deny rules still apply. `--yolo`/`--auto` cannot combine with `-p` — and aren't needed.
- stdout is assistant text with a `• ` transcript prefix (strip if parsing); thinking and
  tool progress go to stderr — keep `2>/dev/null`, drop only to debug a failing run
- run long jobs with Bash `run_in_background`; read the output file on completion
- only delegate from repos without prod secrets in reach — headless mode auto-approves
  regular tool calls, there is no more-restricted non-interactive tier
- follow-ups: `(cd  && kimi --continue -p "$(cat "$P2")" > "$SCRATCHPAD/kimi-last.md" 2>/dev/null)`
  — resumes the latest session in that dir, keeps context (verified: remembers prior run)
- parallel independent orders OK: separate repos/dirs, separate output files.
  `--continue` is cwd-filtered, so never run two concurrent sessions in the same dir —
  they'd race on the same "latest session"

## Prompt contract

Kimi starts with ZERO session context. Every prompt: goal, exact repo/paths, constraints,
non-goals, proof expected, output shape. Spec quality decides success.

## Verify (Claude, always)

- `git status -sb` + read the FULL diff; judge like a hostile contributor-PR review
- run the focused tests yourself or demand proof output; Kimi claims are advisory
- iterate via `--continue`; after 2 failed rounds, take over and do it directly
- ship through normal repo protocol (commit/create-pr skills, CI watch) — Claude's hands
  on the merge button, never Kimi's
- substantive results feed a cross-vendor adversarial review (Claude attacks
  Kimi-built work) before merge

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [boringmarketer](https://github.com/boringmarketer)
- **Source:** [boringmarketer/kimi-first](https://github.com/boringmarketer/kimi-first)
- **License:** MIT
- **Homepage:** https://boringmarketing.com

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** yes
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-boringmarketer-kimi-first-kimi-first
- Seller: https://agentstack.voostack.com/s/boringmarketer
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.