Skip to content

Running on other harnesses

AI-DLC is one harness-neutral core rendered onto the CLI you use. The methodology — the phases and stages, the agents, the scopes, the approval gates — is identical on every harness. What differs is the shell: how gates render, how subagents are dispatched, which session events fire, where config lives. Each chapter here covers one harness's install steps, prerequisites, and the handful of behaviours that differ from the neutral methodology.

Onboarding is installed in each harness's native instruction surface:

Harness Onboarding file
Claude Code .claude/CLAUDE.md (full onboarding)
Kiro CLI and Kiro IDE .kiro/steering/aidlc-onboarding.md
Codex CLI .codex/onboarding.md (also injected through developer_instructions in .codex/config.toml when the project is trusted)
Cursor .cursor/rules/aidlc-onboarding.mdc
opencode .aidlc/onboarding.md
GitHub Copilot Root AGENTS.md (full onboarding, including method imports)

Kiro, Kiro IDE, Codex, Cursor, and opencode share a harness-neutral root AGENTS.md block with other installed harnesses. Their native setup stays in the files above. Copilot's full root block remains exclusive; distinct harnesses also need distinct engine directories to coexist.

Install first

The recommended first-run path for every harness is the checksum-verified native installer followed by aidlc config:

tmp="$(mktemp -d)"
curl -fsSL \
  https://github.com/awslabs/aidlc-workflows/releases/latest/download/install.sh \
  -o "$tmp/install.sh"
sh "$tmp/install.sh"
rm -rf "$tmp"
cd your-project
aidlc config

The installer always includes every harness runtime. aidlc config --harness <name> selects the project surface. Host prerequisites still apply: Codex requires the target project to be a Git repository for project hook discovery.

On Windows, download install.ps1 and invoke it as & $installer. See Windows installation for account scope, automatic User PATH registration, and -NoModifyPath.

Pick your harness:

Harness Invoke Chapter
Claude Code /aidlc Covered throughout the User Guide (its examples run on Claude Code); install in Getting Started.
Kiro IDE (1.x) and Kiro CLI (v3 engine) /aidlc Running AI-DLC on Kiro IDE and Kiro CLI — one tree for both surfaces: prerequisites (Opus 4.8), install, hooks, what's different on Kiro.
Kiro CLI (≥ 2.6) /aidlc Running AI-DLC on Kiro CLI — prerequisites, install, what's different on Kiro.
Codex CLI (≥ 0.145.0) $aidlc AI-DLC on Codex CLI — prerequisites, trust pre-seed, Bedrock config, the git-repo requirement.
Cursor /aidlc AI-DLC on Cursor — one tree for the Cursor IDE and CLI, native subagents and skills, the hooks.json adapter, what's different on Cursor.
opencode (≥ 1.17) /aidlc AI-DLC on opencode — the split .aidlc/ + .opencode/ layout, the adapter plugin, what's different on opencode.
GitHub Copilot (CLI ≥ 1.0.74 / VS Code ≥ 1.130) /aidlc AI-DLC on GitHub Copilot — one install for both surfaces, the .github/ merge, folder trust, what's different on Copilot.

AI-DLC on Kiro (IDE or CLI) works best with Claude Opus 4.8, which requires a paid Kiro plan.

For a manual copy, install Bun, download a specific release's aidlc-copy-runtime-X.Y.Z.tar.gz, extract it, and copy the complete runtime/<harness>/ directory; the native aidlc executable is not required. Do not copy generated trees from a repository checkout. Framework developers may instead run bun scripts/package.ts in a source checkout to materialize the ignored local dist/ and dist-release/ outputs. Each harness chapter keeps the manual-copy instructions under a clearly labeled alternative.

After aidlc update, run aidlc doctor to see project/runtime version skew and refresh each project with aidlc config between workflows. Config refuses an active-workflow refresh, protecting running work from changed stage or graph definitions.

This set is open: a new harness gets its own chapter here, added from the same template. For building a new harness (the source contract — manifest, hook adapter, emit.ts), see the Harness Engineer Guide's Porting to a New Harness.

Whichever harness you run, the methodology is the same — start with Your First Workflow and the Phases and Stages tour.