Skip to content

SETUP.md

Latest commit

 

History

History
99 lines (61 loc) · 3.95 KB

SETUP.md

File metadata and controls

99 lines (61 loc) · 3.95 KB

AI Tool Setup

Before invoking the agent, copy the relevant folder from this repo into the root of your repo/project. The agent file must be present at the expected path for your AI tool to detect it.

Each tool has two ways to run the pipeline:

  • Automated (recommended) — copy the runner script from scripts/ into your repo and run it. The script executes stages 1 → 1.5 → 2 → 3 → 3.5 → 4 in sequence with model switching between generation and verification stages, each in a fresh CLI session.
  • Manual — invoke the agent yourself, one stage at a time. Use a fresh session for every verification stage (1.5, 3.5, 4).

Jump to: Claude Code · Gemini CLI · GitHub Copilot · Codex · Cursor

Claude Code

Install Claude Code. Copy .claude/agents/ into your repo root.

Automated (recommended): Copy scripts/acf-run-claude.sh into your repo root and run it:

./acf-run-claude.sh

Manual — interactive: use acf-context-agent in the chat interface, then Run Stage 1: Onboard.

Manual — CLI: claude --agent acf-context-agent -p "Run Stage 1: Onboard"

Gemini CLI

Install Gemini CLI. Copy .gemini/agents/ into your repo root.

Automated (recommended): Copy scripts/acf-run-gemini.sh into your repo root and run it:

./acf-run-gemini.sh

Manual: Enter Use 'acf-context-agent' and Run Stage 1: Onboard.

GitHub Copilot

Copy .github/agents/ into your repo root.

Automated (recommended): Copy scripts/acf-run-copilot.sh into your repo root and run it:

./acf-run-copilot.sh

Manual — VS Code: Install the GitHub Copilot and GitHub Copilot Chat extensions. In the chat panel, select acf-context-agent from the agent list.

Manual — CLI: Install GitHub Copilot in the CLI. Invoke the agent using the acf-context-agent name.

Codex

Install Codex CLI. Copy .codex/agents/ into your repo root.

Automated (recommended): Copy scripts/acf-run-codex.sh into your repo root and run it:

./acf-run-codex.sh

Manual: Enter Use 'acf-context-agent' and Run Stage 1: Onboard.

Cursor

Install Cursor. Copy .cursor/agents/ into your repo root.

No runner script is provided — Cursor's chat interface is the supported entrypoint. Select acf-context-agent from the agent list, then invoke each stage manually starting with Run Stage 1: Onboard. Start a new chat for every verification stage (1.5, 3.5, 4).

Runner scripts — defaults & overrides

Tool Script Generate model Verify/Review model
Claude Code scripts/acf-run-claude.sh claude-sonnet-4-6 claude-opus-4-6
Codex scripts/acf-run-codex.sh gpt-5.4 gpt-5.4
GitHub Copilot scripts/acf-run-copilot.sh gpt-5.4 claude-opus-4.6
Gemini CLI scripts/acf-run-gemini.sh gemini-2.5-pro gemini-2.5-pro

All models are overridable via environment variables, e.g.:

GENERATE_MODEL=claude-sonnet-4-6 VERIFY_MODEL=claude-opus-4-6 ./acf-run-copilot.sh

Notes:

  • Run scripts from your target repo root, not from the ACF repo. Each script expects the corresponding agent folder (.claude/agents/, .gemini/agents/, etc.) to exist in the current working directory.
  • Stage 5 (Update) is not part of the runner scripts. Run it separately on a schedule, after significant releases, to detect drift.
  • If a stage exits non-zero, the script logs a warning and continues. Re-run individual stages manually if needed.

Windows: The runner scripts are bash and require WSL2 or Git Bash. If neither is available, use the manual invocation steps for your tool — every stage can be triggered by typing the prompt into the chat. The agent specs, generated docs, and CLI tools themselves are fully cross-platform.