The AI-First Taskforce's shared body of knowledge for building production software AI-first. It's captured in two complementary forms:
- Prose — best practices,
CLAUDE.mdtemplates, trial artifacts, and the AI-First Engineering Strategy. - Skills — the
aif-*Claude Code skills underskills/: executable workflows and a form of documentation, each encoding a best practice precisely enough for an agent to carry it out.
The skills are delivered as a gh CLI extension that installs them into your own Claude Code (~/.claude/skills/) — see Installation.
Install as a gh CLI extension:
gh extension install String-sg/gh-ai-first-taskforceThen run setup to install the taskforce's Claude Code skills into ~/.claude/skills/:
gh ai-first-taskforce setupSkills are available to Claude Code automatically once installed. To get the latest skills after an extension update:
gh extension upgrade gh-ai-first-taskforce && gh ai-first-taskforce setupThe AI-First Taskforce aims to increase developer productivity through practical application of generative AI in software engineering workflows.
See AI-First Engineering Strategy — how AI modernizes software engineering: the progression from ad-hoc AI use to structured agentic software engineering, and how teams adopt it safely.
| Project | Description |
|---|---|
| Templatized skills (this repo) | Reusable Claude Code skills and CLAUDE.md templates that any team can adopt, derived from real project trials. |
| Personal data detection | Tooling to detect personal data in codebases and datasets before they reach production or external services. |
Local codebase sensitivity scanner
Scan a local codebase using a local LLM — no data leaves the device — to determine its sensitivity and information classification. The output determines which deployment environment the project is eligible for (Greenlane OPEN, OFFICIAL, or OFFICIAL-CLOSED), without risking data exposure during the scan itself.
Data masking for restricted projects
For projects that cannot be deployed to commercial or cloud LLM environments, explore an LLM-assisted pipeline to mask sensitive data and copywriting before it reaches an external model — expanding the scope of projects that can benefit from AI tooling without compromising data handling requirements.
gh-ai-first-taskforce/
├── gh-ai-first-taskforce Extension entry point — `gh ai-first-taskforce setup`
├── lefthook.yml Lefthook config — pre-commit secret scan, pre-push main protection
├── hooks/ Git hook scripts wired up by Lefthook
├── skills/ Claude Code skills (aif-*) installed by the extension
│ ├── README.md Catalogue of installed skills
│ ├── aif-code-review/
│ ├── aif-create-issue/
│ ├── aif-git-hooks-setup/
│ ├── aif-implement-issue/
│ ├── aif-lint-setup/
│ ├── aif-split-issue/
│ └── aif-update-npm-dependencies/
├── docs/
│ └── ai-first-engineering-strategy.md How AI modernizes engineering — the progression and toolkit
├── templates/
│ ├── CLAUDE.md Generalized CLAUDE.md — copy to a new project before build week
│ ├── trial-review.md Blank post-trial review template
│ ├── trial-goals.md Goals and success criteria template — fill in before each trial
│ └── skills/ Legacy: copy-in stack/function review skills (superseded by aif-* skills)
└── trials/
└── sums/ Artifacts from Trial 1: SuMS (Feb–Mar 2026)
├── CLAUDE.md SuMS project rules (the source for templates/CLAUDE.md)
├── trial-review.md SuMS post-trial review with gaps log
└── skills/ SuMS-specific skills (source for templates/skills/by-stack/nextjs-ts-prisma/)
The taskforce ships Claude Code skills as a gh extension. Install them with gh ai-first-taskforce setup (see Installation) — they land in ~/.claude/skills/ and Claude Code picks them up automatically. Each skill is self-describing: its SKILL.md declares when to trigger, so there is no router to maintain.
See skills/README.md for the catalogue of installed skills — the single source of truth, kept in sync as skills are added or removed.
For the strategy behind these skills — how teams adopt them, and how the toolkit itself is built — see AI-First Engineering Strategy.
Before the gh extension, review skills were distributed as copy-in templates under templates/skills/, organized by stack and by function with a SKILLS.md routing index (pre-merge-audit and review-pr for Next.js · TypeScript · Prisma). These are superseded by the aif-* skills above and kept for reference. To use one, copy its directory into a project's .claude/skills/ and commit it.
Before the build week:
- Fill out
templates/trial-goals.mdwith the project context, what you want to learn, and success criteria. Sign off with the PM, SWE, and DevOps before the build starts. - Copy
templates/CLAUDE.mdto the new project repo asCLAUDE.md. Edit all[ ]placeholders for the project's stack, hosting, and environments. - Work through the New Project Init Checklist inside that
CLAUDE.mdbefore any application code is written. - Install the taskforce skills with
gh ai-first-taskforce setupso theaif-*skills are available from day one.
During the trial:
- SWE reviews Claude's commits on an agreed cadence (daily async is the baseline). Running
aif-code-reviewat the end of each session is a lightweight way to surface violations before they accumulate. Log gaps as they emerge — don't wait for the end.
Before merging any PR:
- SWE runs
aif-code-reviewon the PR branch to scan for violations and capture findings before merge. Only merge once the findings are resolved.
After the trial:
- Fill out
templates/trial-review.md. Add the completed review totrials/<project-name>/trial-review.md. - Review the gaps log and update
templates/CLAUDE.mdwith any new rules that would have prevented them. Update the skills if new automated checks are warranted.
See CONTRIBUTING.md for the full guide — ways to contribute, the issue-first workflow, local setup (installing the git hooks), branching and Conventional Commits, how to add a skill and the quality bar, and the PR flow.
| Trial | Project | Period | Review |
|---|---|---|---|
| 1 | SuMS | Feb–Mar 2026 | Review |