Skip to content

neekolas/code-factory

v0.4.3MIT

Skills for planning, implementation, review, and repository care

adversarial-review

Run one adversarial review pass on one commit and report verified findings. Use when the user asks for an adversarial review of a branch, commit, or PR, or when an orchestrator runs a pass.

audit-tests

Judge whether tests detect real regressions. Use for focused test review, new test design, or a scoped audit of weak, brittle, or duplicate tests. Preserve independent contracts and report evidence before cleanup.

babysit-pr

Use when taking one feedback round on a pull request or stack: collect current-head failures and review feedback, apply verified fixes in stack order, and push once. Repeat rounds until the PRs are ready. Works with Graphite (gt), GitHub stacked PRs (gh stack), and standalone branches.

boat

Create, monitor, steer, resume, and stop coding tasks on Boat VMs. Configure a local CLIProxyAPI gateway over Tailscale and prepare Boat snapshots for Claude Code and Codex.

calling-claude

Run Claude models from a non-Claude agent such as Codex. Claude Code sessions use native subagents instead.

calling-codex

Run Codex models from a non-Codex agent such as Claude Code. Codex sessions use native subagents instead.

model-choice

Default models and effort levels for implementation, review, chores, code scanning, and research. Use when selecting models for agent work.

orchestrate-execution

Guide native orchestration of implementation, code review, and PR feedback with a few rules. Use only when the user explicitly invokes orchestrate-execution.

security-audit

Run a full security audit of one commit - parallel hunting lanes across models, dedup, exploit validation, and a remediation plan. Use only when the user explicitly invokes it.

session-retro

Run only when the user asks for a retro of a finished orchestrated run - mines the orchestrator's and subagents' transcripts, groups findings with earlier retro proposals from any agent, proposes repository script, check, skill, and AGENTS.md changes with tokens-saved and time-saved estimates, puts rule and skill text in a review document, and turns approved proposals into fix PRs or issues

writing-plans

Use before a change that needs more than one commit, touches a public contract, or needs a decision from the user - writes a plan whose requirements use the repository's spec row format, gets a clean-context adversarial review, and stops at user approval. Replaces code-factory:writing-specs