XIII.Installation book · v0.19

Codex / Claude Code · Windows / macOS / Linux

One prompt. Your agent installs STOLZ.

Copy the prompt into Codex or Claude Code. The same request works on Windows, macOS, and Linux.

Codex / Claude Code

Install all STOLZ A.I. skills and their reference files for yourself from https://github.com/Sergey360/stolz-ai/releases/latest. Verify the archive's SHA-256, use your user-level skills directory, preserve existing skills, and verify the installation.

Paste this into your agent's chat. It can download and install the skills for your system.

After installation, open a new agent session.

Manual installation and team setup

1. Verify the archive

Download the package and compare its SHA-256 before installing it. The canonical archive contains 214 files.

SHA-256cdde8bad689c9247c7b1ab406f193f876ebb4f8e8694d0dff94055b4529d501a

Download the package Browse tagged source

2. Install in the consumer project

Install the downloaded archive as a development dependency. The CLI then runs without fetching another package.

terminal / npm

npm install --save-dev /absolute/downloads/stolz-ai-0.19.0.tgz

3. Resolve, preview, then install

Resolve the Codex profile, inspect the deterministic installation plan, and add `--apply` only after reviewing the destination and owned files.

terminal / profile

npx --no-install stolz-profile resolve --runtime codex
npx --no-install stolz-profile install --runtime codex --destination /absolute/path/to/skills --dry-run
npx --no-install stolz-profile install --apply --runtime codex --destination /absolute/path/to/skills

4. Run doctor

Doctor makes no provider or model call. It reports installation state, the observed runtime version, compatibility basis, lazy-adapter availability, and the next action.

terminal / doctor

npx --no-install stolz-profile doctor --runtime codex --runtime-version <version> --destination /absolute/path/to/skills

5. Commit a team lock and verify it in CI

The lock contains package and profile identity, not credentials or local paths. `verify-lock` exits non-zero on drift, an invalid lock, or an incompatible reported runtime version.

terminal / team lock

npx --no-install stolz-profile lock --apply --runtime codex --lockfile ./stolz-profile.lock.json
npx --no-install stolz-profile verify-lock --runtime codex --lockfile ./stolz-profile.lock.json

6. Review an update before applying it

The first command prints the future change. Apply only after review; locally changed or missing STOLZ-owned files stop the update. The v0.18.0 → v0.19.0 update, recovery, conflict, and rollback paths passed package smoke.

terminal / update

npx --no-install stolz-profile update --runtime codex --destination /absolute/path/to/skills
npx --no-install stolz-profile update --apply --runtime codex --destination /absolute/path/to/skills

7. Recover an interrupted update

If doctor reports `recovery_required`, preview recovery first. Apply restores the verified prior installation recorded by the transaction journal.

terminal / recover

npx --no-install stolz-profile recover --dry-run --destination /absolute/path/to/skills
npx --no-install stolz-profile recover --apply --destination /absolute/path/to/skills

8. Roll back with the same discipline

Preview rollback before applying it. Rollback refuses to overwrite a locally changed owned file.

terminal / rollback

npx --no-install stolz-profile rollback --dry-run --destination /absolute/path/to/skills
npx --no-install stolz-profile rollback --apply --destination /absolute/path/to/skills

Optional: install stolz-browser separately

Copy the complete browser directory with both reference files. It has no dependency on a personal recovery skill or sibling adapter.

terminal / optional skill

mkdir -p .agents/skills
cp -R node_modules/stolz-ai/skills/stolz-browser .agents/skills/

Use it for controlled tabs and recovery decisions. Preserve the selected browser and required UI checks; installing instructions does not supply browser tools or establish crash prevention.

Optional: install stolz-evidence separately

Copy the full evidence directory with its references. Profiles manage five optimization skills; update or remove the optional skill separately.

terminal / optional skill

mkdir -p .agents/skills
cp -R node_modules/stolz-ai/skills/stolz-evidence .agents/skills/

For an optional short project rule, review the dry run before adding --apply. Select codex, claude-code or qwen-code and an existing absolute project directory. Verify effective instructions in your runtime; file merging alone does not establish discovery.

terminal / project rule dry run

node node_modules/stolz-ai/tools/evidence-instructions.mjs --project-root /absolute/project --runtime codex

Optional: install stolz-guard separately

Profiles install five optimization skills. Copy the complete guard directory for instruction security review, including its references. Run this command once for a new destination.

terminal / optional skill

mkdir -p .agents/skills
cp -R node_modules/stolz-ai/skills/stolz-guard .agents/skills/

The profile lock, status, update and uninstall manage only profile-owned files. Review and update the separate guard installation yourself. The host enforces tool permissions.

Optional: install stolz-mcp separately

Copy the complete MCP skill with both references. The Node package also supplies the offline stolz-mcp CLI and the stolz-ai/mcp-code API.

terminal / optional skill

mkdir -p .agents/skills
cp -R node_modules/stolz-ai/skills/stolz-mcp .agents/skills/
npx --no-install stolz-mcp --help

No MCP settings are changed. The host supplies the transport and authorized execution environment; code execution is not a sandbox or permission bypass.

§ 3scope / evidence

What these checks prove

They prove package identity, selected configuration, and manifest-owned files. They do not prove model compatibility, provider-native tokens, billing, cost, or savings.

Real-project follow-up

The status remains `followup_gap`: no named real-pilot owner or 28-day real evidence window is recorded yet.

Return to the v0.19 release