forge

Audit comments and god files with eslint-plugin-code-quality and report actionable findings. Use when the user asks to review comment quality, find historical narration, reduce comment density, or find oversized files, long functions, and crowded directories.

Run one wave of delegated issue work: triage what is open, group what shares a place, dispatch each through a shipped role, and fold what came back. Invoke when several issues are to be worked by delegated runs rather than by this session, or when a wave that was dispatched has reported and owes its fold. For taking one issue to landed code yourself, use the issue-flow skill instead; for reading or filing without either, the forge skill. Triggers on "dispatch a wave", "what should we work next", "delegate these issues", "fold the wave", "triage the backlog".

Drive a Forge issue tracker from the terminal with the `forge` CLI — browse, read, file, comment on and attach to issues without an MCP client connected. Invoke for any task that reads or writes the backlog: listing open issues, filing a defect, posting a finding, recording a dependency, or reading an issue body before working it. The same CLI carries `forge cloudflare` for zones, DNS records and cache purges. It carries `forge coolify` for a pinned project's deployments. It carries `forge codex` for a review of this turn's documents by a model on another provider. Triggers on Forge, tracker, backlog, issue, ISS-nn, "file an issue", "what's open". Also on Cloudflare, DNS record, zone, purge cache. Also on Coolify, deploy, redeploy, "restart the app", "is it running". Also on codex, "second opinion", "review this plan".

Profile a project's gate — the check, lint, build and test pipeline a change has to pass — and make the same gate answer faster on the same machine without letting anything through it. Use when a gate run has grown into a tax every run pays, when its whole-run time crosses the budget the project set or has grown by a quarter since it was last profiled, or when asked why the pipeline takes as long as it does. Triggers on "the gate is slow", "profile the gate", "speed up the suite", "why does check take so long", "run the tests in parallel", "the fixtures are piling up".

Judge the harness by its own numbers: run the evals that compare the last window of consults and of issue-flow runs with the window before, attribute each figure that moved to the change that separates the windows, and file what the numbers say against the issue that made the change. Invoke when a release or a consult prints the line naming an eval, or before a harness change is judged by the feel of the next few runs. For building a change the numbers point at, use the issue-flow skill; for a wave of issues, the dispatch skill. Triggers on "run the eval", "did that release make runs cheaper", "judge the harness", "read the eval".

Implement a tracker issue end to end — plan it, build it, prove it, ship it. Invoke when an issue has to become deployed code: verifying what it claims, writing the plan the tracker holds, implementing, proving by running, and taking it to done. For reading, listing or filing issues without implementing them, use the forge skill instead. Triggers on "work ISS-nn", "implement this issue", "fix ISS-nn", "làm issue", "xử lý ISS-nn", "ship this issue".

Drain the issues a project's independent judgement has left standing at `developed`: claim each, read whether anybody has to witness it at the running product, judge it or dispatch a judging run for it, take it as far as its record earns, and stop when the status reads empty. Invoke when a project that declared an independent judgement has work waiting at that status, or when a judging run has to be given its own method. For a wave of open issues taken by building runs, use the dispatch skill instead; for taking one issue from its title to landed code, the issue-flow skill. Triggers on "drain developed", "run the QA master", "judge what is waiting", "who judges this", "the judging queue", "nothing has judged these".

Drive one release job from a change that has landed to a verified production deployment and the account of what happened — establishing the state before changing it, repairing forward where the world does not match the plan, and stopping at a declared bound rather than at a hunch. Invoke when a release run names this method, when a promotion or deploy did not come up as intended, or when a release has to be taken from a landed commit to a deployment somebody has read. For taking one tracker issue from its title to landed code, use the issue-flow skill instead. Triggers on "run the release", "promote to production", "the deploy did not come up", "the release job is stuck".

Set up eslint-plugin-code-quality, its limits, and its Claude Code edit hook in the current Node project. Use when the hook reports a missing plugin or configuration, when a project has no ESLint at all, or when the user asks to install or configure code-quality linting.

Use whenever Vietnamese text has to be produced or judged — translating an i18n locale file (en.json → vi.json), localizing UI strings, writing or translating Vietnamese documentation and README files, adding a vi locale to a project, or reviewing existing tiếng Việt that reads like machine translation. Triggers on Vietnamese, tiếng Việt, vi.json, vi-VN, locale, i18n, l10n, dịch, bản địa hóa, "translate to Vietnamese". Provides the `vi-natural` CLI.