Setup and

AND Setup

Audit and configure one repository to run the AND delivery loop with GitHub as its workflow-state authority.

Setup Contract

Repository setup is conformant when:

• the repository satisfies Read Setup Readiness from setup-readiness.md;
• every directory containing Agent instructions follows the Agent Instruction Policy below;
• no observed GitHub workflow-state contradiction makes stage, relationship, ownership, or lifecycle meaning ambiguous.

External skill readiness and the next-skill receipt are report results rather than repository conformance conditions. When the user explicitly requests full readiness, missing external skills make that result incomplete while repository conformance remains separate.

Process

1. Explore Current State

Work outward from observable evidence:

• Resolve the target repository from an explicit target, current checkout, and Git remotes. More than one plausible repository is a decision, and no GitHub repository is a setup blocker.
• Use and-workflow-contract and load setup-readiness.md for Read Setup Readiness: inspect Issues availability, required labels, issue-write and self-assignment permission, and native relationship capability. Load wayfinding.md when Wayfinding is installed.
• Read relationship-api.md before interpreting native relationship responses or planning a reversible capability probe.
• Inspect existing GitHub workflow state for contradictions that a setup write must preserve and route to and-sweep rather than overwrite. An unresolved contradiction keeps repository conformance incomplete.
• Scan every AGENTS.md and CLAUDE.md outside dependency, generated, and VCS directories. Classify each containing directory, resolve symlinks, and follow references only far enough to determine effective workflow rules.
• Check and-workflow-contract, and-interview-contract, grilling, research, prototype, tdd, and code-review separately as Environment Readiness.

The audit is read-only. A write capability that cannot be proven from authoritative permission and role evidence remains unverified until an approved envelope authorizes a reversible probe.

Completion criterion: every Setup Contract condition has evidence, a gap, or one named unresolved decision, and environment readiness is known.

2. Resolve Genuine Decisions

Investigate facts before asking. GitHub is the required workflow authority.

When a user decision remains:

• ask one decision at a time in dependency order;
• give valid options, recommendation, rationale, and impact;
• use the answer to finish dependent evidence gathering before asking again.

Genuine decisions include choosing among multiple plausible repositories, merging non-identical paired Agent instructions, normalizing a mixed instruction layout without an existing policy, and authorizing a reversible capability probe. Missing labels and a missing or stale root AND section are ordinary setup gaps.

Completion criterion: no unresolved choice can materially change the proposed writes.

3. Prepare One Write Envelope

Report the assessment at the requested depth:

• default to a compact satisfied summary plus every real gap;
• return a concise no-op for a conformant repository;
• show every condition and its evidence only for an explicit full audit;
• stop without writes for audit-only requests.

For setup or repair, present one envelope containing:

• every Agent-instruction file and section to create, replace, move, or delete;
• every GitHub label or approved reversible capability-probe action;
• exact label names and the complete proposed replacement for each user-readable instruction section;
• the reason and user-visible impact of each action.

Show a full patch only when requested. One approval authorizes the complete envelope. Any later action outside or contrary to it requires a new envelope.

Completion criterion: the user can authorize all necessary writes once, with no hidden or per-surface follow-up confirmation.

4. Apply The Approved Minimum

• Perform only approved writes and GitHub actions.
• Preserve unrelated instructions, documentation, issues, work records, relationships, ownership, and implementation artifacts.
• Leave satisfied GitHub state unchanged and remove every disposable probe artifact before declaring its capability verified.
• Create project-specific documentation only for real repository facts unavailable from the AND skills and workflow contract.
• Install external skills only when explicitly requested and included in the envelope.

Completion criterion: every approved gap is addressed, every probe is cleaned up, and nothing outside the envelope changed.

5. Re-audit And Report

Run the same assessment against the same repository after writes.

• Verify every Setup Contract condition from current evidence.
• Verify that a second apply would be a no-op.
• Report repository conformance separately from environment and full readiness.
• Name changed files and GitHub actions, remaining blockers, missing-skill install commands, and exactly one next workflow skill.
• Route back to setup-and rather than downstream when any conformance condition remains unsatisfied.

Completion criterion: the receipt proves current state, idempotence, and one valid next route without making the receipt itself a setup requirement.

Setup Outcomes

Starting state	Required result
Fresh GitHub repository	Verify capabilities, propose required labels and root instructions in one envelope, apply when approved, then re-audit.
Existing conformant setup	Make no writes and return the compact conformant receipt.
Incomplete setup	Preserve satisfied state and include only proven gaps in the envelope.
Missing or ambiguous GitHub repository	Name the exact repository boundary; resolve ambiguity or GitHub availability before proposing workflow writes.
Capability failed or unverified	Name the failed evidence and impact, then offer a capability remedy or approved reversible probe; readiness remains incomplete.
Audit only	Return evidence and gaps, then stop without mutation.

Agent Instruction Policy

Apply this policy directly:

• AGENTS.md holds shared cross-Agent instructions.
• A sibling CLAUDE.md imports @AGENTS.md first, then may append Claude Code-specific rules. Shared instructions appear only in AGENTS.md.
• If only AGENTS.md exists, preserve it and add the sibling CLAUDE.md adapter.
• If only CLAUDE.md exists, move shared content to sibling AGENTS.md, then replace it with the adapter. Use git mv for tracked files when possible.
• Leave a conformant pair unchanged.
• For a pair with non-identical or ambiguously shared content, obtain one merge-policy decision before mutation.
• For a mixed repository layout without an explicit policy, report each directory class and obtain one normalization-policy decision.
• If no Agent instruction files exist, propose root AGENTS.md plus the adapter without asking the user to choose a filename.

The adapter is:

@AGENTS.md

The root AND section is:

## AI-native development

This repository uses the AND delivery loop.

- Use GitHub Issues, labels, native relationships, comments, and assignees as workflow state.
- Use `ask-andie` when the next workflow skill is unclear.
- Use the installed `ai-native-development` skills and `and-workflow-contract` for workflow rules.

Environment Readiness

Check and-workflow-contract, and-interview-contract, grilling, research, prototype, tdd, and code-review from the session skill list or documented skill-list command. Report unverified when neither establishes availability.

Missing skills do not authorize installation. When installation is requested, use the owning repository and known Agent targets:

npx --yes skills add Andiedie/skills -g --skill and-workflow-contract and-interview-contract --agent <known-target...>
npx --yes skills add mattpocock/skills -g --skill <missing-matt-skill...> --agent <known-target...>

When targets are unknown, omit --agent for interactive selection. Keep skills from different repositories in separate commands.

Report Shape

After a successful apply, report:

Repository setup: conformant
Workflow state: GitHub ready
Changed: <files and GitHub actions, or none>
Environment readiness: <ready | missing or unverified skills>
Full readiness: <ready | incomplete, only when requested>
Next: <one AND workflow skill>

Before writes, list the target repository, exact file actions, exact GitHub actions, environment readiness, and any blocker. Omit empty groups and passed-condition inventories from the default report.

Boundaries

• Setup writes only approved minimum GitHub state, Agent instruction integration, repository-specific setup facts, and explicitly authorized skill installation.
• Existing work records, product requirements, implementation artifacts, and workflow meaning remain unchanged.
• Delivery-loop stage work begins only after repository setup is conformant; setup returns one next-skill route without claiming, implementing, closing, merging, or releasing delivery work.