standardize-project

Audit and Standardize a Project

Inspect first, propose a plan, then change. Never overwrite existing project configuration blindly — the goal is a repo that meets the standard and still works the way its team expects.

For a read-only mapping report (which guides apply, exceptions vs gaps, no migration), use standards-map instead — or run it first, then return here to apply.

Target

Resolve the target repository before doing anything. If none was given, ask and stop. Use explicit paths for every read and edit — it is easy to audit one repo and accidentally edit another.

Source of truth

The standards live in The-Focus-AI/standards. Read these before recommending anything: AGENTS.md, best-practices/GDE-009-technology-defaults.md, best-practices/GDE-008-skills.md, STD-007 and STD-008 for secrets and deployment, and whichever deployment guide applies — best-practices/GDE-010-vercel-deployment.md, best-practices/GDE-005-gcp-deployment.md, or best-practices/GDE-004-gce-gaia-runtime.md. Also best-practices/GDE-002-clerk.md, best-practices/GDE-007-pi-extensions.md, best-practices/GDE-003-fnox-secrets.md, and the default-project template for missing-file examples only.

If the standards conflict with the target project's own instructions, name the conflict and ask — do not silently pick a side.

Audit order

1. Agent instructions — AGENTS.md, CLAUDE.md, .cursor/rules, equivalents.
2. Tooling — mise.toml, .tool-versions, package manager files, lockfiles, scripts.
3. Secrets — .env, .env.*, fnox.toml, .fnox/, CI secret references, README setup steps.
4. pi setup — .pi/settings.json, packages, local session/cache ignores.
5. Skills — skills-lock.json, .agents/skills/, .pi/skills/.
6. Workflow — README, docs, GitHub remote, tracker usage, spec/ticket docs, wayfinder maps.
7. Git hygiene — .gitignore, committed generated files, committed-secret risk.

Required output before any change

1. Current state — what exists, what is missing.
2. Compliance gaps — each mapped to the standards file that requires it.
3. Proposed diff — exact files to create or edit, with the intended change.
4. Secrets migration plan — discovered key names only, never values, plus which 1Password vault and service account to use.
5. Commands to run from the target project.
6. Questions and approvals needed — especially before touching secrets or deploys.

Do not apply changes until the user approves, unless they asked for immediate execution.

Applying the migration

mise

Add tools with mise use, never by hand-editing:

mise use node@22 npm:pnpm fnox gh ripgrep fd
mise use npm:@earendil-works/pi-coding-agent
mise use npm:markdownlint-cli2

Merge tasks additively and preserve project-specific ones. Required where applicable: install, setup, dev, lint, test, deploy, secrets:check, secrets:list. Add [env] with _.file = ".fnox/env" if missing.

fnox and 1Password

Follow best-practices/GDE-003-fnox-secrets.md. Target end state:

• A dedicated 1Password vault per project.
• A scoped service account with access to that vault only.
• .fnox/env holding OP_SERVICE_ACCOUNT_TOKEN and nothing else beyond bootstrap values.
• fnox.toml mapping environment variables to 1Password items.
• Plaintext env files quarantined under .fnox/migrated-env/ or replaced with .env.example, after confirmation.
• .fnox/, .env, .env.* gitignored — except .env.example.

Never print, commit, or paste a secret value. Key names only, in the plan and in conversation.

Skills

Install missing skills with skills add (mise-pinned); never copy skill directories by hand. Prefer --skill <folder> over adding a whole upstream repo. The canonical set is best-practices/GDE-008-skills.md and the locked skills-lock.json.

Remove deprecated to-prd and to-issues entries when migrating. Replace title-case lock keys with kebab-case. Ensure the planning workflow skills are present and referenced in the project's agent instructions.

pi

Merge .pi/settings.json — project-local sessions plus the required packages. Do not delete project-specific pi config.

Agent instructions

Update or create the target's AGENTS.md additively. It should tell a future agent where to read project context, that tooling goes through mise, that secrets go through fnox and 1Password, how the planning workflow runs, how issues are tracked, and which commands prove the project is healthy.

Verification

From the target project, run what is available:

mise trust && mise install && mise tasks
mise run install && mise run lint && mise run test
fnox check

If a command cannot run, say why and what remains. An audit that ends with unverified claims is half an audit.

Pitfalls

• Inspect before proposing. A plan written from assumptions about the repo wastes the review it asks for.
• Additive, not replacing. Project-specific tasks and pi settings are there for a reason.
• Secrets are the dangerous part. Get explicit approval before moving them, and verify the project still starts afterward.
• A gap you cannot fix is still a finding. Report it rather than quietly dropping it.