Bootstrap new projects with the Harness Engineering skeleton and Claude Code binding.
npx skills add https://github.com/Alan-IFT/harness-kit --skill harness-init-alan-ift
Please help me install this Agent Skill. Skill: harness-init Source: https://github.com/Alan-IFT/harness-kit/tree/main/skills/harness-init Command: npx skills add https://github.com/Alan-IFT/harness-kit --skill harness-init-alan-ift
.harness/ rules, playbooks, scripts, and docs. - Interactive configuration: Asks six questions (project type, stack, hooks, partitioning, output language, AI customization) and substitutes placeholders like {{PROJECT_NAME}} and {{STACK}} across all .tmpl files. - Optional AI-native customization: Drafts a tailored .harness/rules/50-<slug>.md grounded in your manifests (package.json, go.mod, etc.) with source citations, validated against four invariants with static-stub fallback. - Use Case: You create an empty repo for a new FastAPI backend. Run the skill, answer the six questions, and it produces the full .harness/ tree, CLAUDE.md stub, verify_all gate, safety hooks, and git pre-commit hook — ready for /harness pipeline work. ## Quick Start Ask the AI to initialize Harness Engineering in the current empty project directory and answer the setup questions about project type and stack.High-intent search queries and answers about installing and using this skill.
Run harness-init in the empty project directory. It asks six questions about project type, stack, hooks, partitioning, and language, then copies the full .harness/ skeleton, CLAUDE.md stub, verify_all scripts, and safety hooks so the 7-agent pipeline works immediately.
harness-init is for new empty projects or greenfield directories with no existing code. harness-adopt is for existing codebases that already contain code. Running harness-init in a directory with .harness/, .claude/agents/, or CLAUDE.md stops with a warning to avoid overwriting files.
Yes. It offers three project types: Fullstack, Backend/API service, and Generic (CLI, library, mobile, ML, embedded). Each type copies a matching template overlay with appropriate rules, partition agents, and verify_all presets.
Yes, via the opt-in step 5b. The skill reads your stack description, top-level files, and manifests like package.json or go.mod, then drafts a tailored 50-<slug>.md rule file with source citations. Output is validated against four invariants and falls back to the static stub on failure.
It can install a Claude Code Stop hook and a git pre-commit hook that keep generated files in sync with .harness/ edits. Both are optional via question 3; the destructive-command guard-rm PreToolUse hook is always wired for safety.
Placeholder substitution depends on the hook-spec.sh script answering command values for your shell. If the spec is missing, exits non-zero, or returns empty, the skill stops and leaves the token unresolved rather than improvising a command. Re-copy the scripts from templates/common/.harness/scripts/.
yes506Generate standardized project baselines with directory structures, build configurations, and CI stubs.
MaxWolf-01Runs a structured interview session that sharpens plans and produces glossary and ADR documentation.
cbrostromEvaluates external tools and ideas against an existing agent setup with adopt, borrow, or reject verdicts.
thepradipPackage, template, and deploy Kubernetes resources as Helm charts.