Roll out the canonical Aether audit capability across the repository fleet

[szmyty]

Outcome

Adopt the canonical Aether repository-audit capability across applicable Ego Hygiene repositories through Pace's reviewable fleet-convergence model, without copying/forking audit specs, skills, or agents into 29 independently maintained implementations.

The goal is that a fresh AI conversation, Copilot session, local model, or approved automation can enter an applicable repository and discover the same pinned audit capability and local invocation contract.

Dependencies

Do not begin fleet writes until the upstream artifacts are reviewed and releasable:

• composable audit profiles + mature skill: egohygiene/aether#86;
• applicable Aether distribution/install mechanism;
• organization policy/profile declaration from Hygiene where required;
• validation/evidence hooks from Egolint/Relay as applicable;
• Observatory adoption/read-model support;
• Holon materialization contract for new repositories when needed.

Use immutable released/pinned artifacts, not mutable sibling default branches.

Desired state

For each applicable repository, Pace should be able to determine:

• whether the canonical audit bundle/capability is applicable;
• selected Aether distribution/profile version;
• whether the repository has the required local pointer/config/projection;
• whether provider-specific projections are current;
• whether the canonical audit output location is available/configured;
• whether local instructions point to the managed capability rather than copied policy;
• whether the repository has a reviewed exception;
• whether the installed state is current, missing, drifted, stale, incompatible, or blocked.

Installation model

Prefer the smallest repository-local footprint possible.

Repository-local files should primarily contain:

• a version/profile pin or managed capability declaration;
• generated provider projection/pointer only where the host requires it;
• repository-specific configuration such as applicability, exclusions, or local evidence paths.

Do not copy the full Aether auditor specification or reusable skill prose into every repository as hand-maintained source.

Rollout protocol

1. Observe the current fleet and classify existing audit implementations.
2. Produce a no-write plan showing exact per-repository changes and exceptions.
3. Reconcile legacy/custom auditor files non-destructively; preserve useful repository-specific configuration while removing duplicate canonical semantics only after review.
4. Pilot on the contract-owning repositories plus materially different canaries.
5. Open one bounded PR per repository by default.
6. Validate the installed capability and provider projections locally/CI where applicable.
7. Re-observe after merges and update the desired-state lock.
8. Advance from observe/advisory toward stronger enforcement only after adoption evidence supports it.

Provider-neutral requirement

The installed capability must not assume one execution host.

A repository should be able to expose the same Aether audit semantics to:

• GitHub Copilot/custom agents;
• ChatGPT/manual fresh-conversation workflows;
• local/offline agent runners;
• future provider adapters;

without each provider becoming an independent source of audit policy.

Existing/legacy audit handling

Classify repository-local audit material as one of:

• canonical Aether-managed projection;
• repository-specific extension/configuration;
• historical report/evidence;
• legacy duplicate requiring migration;
• unrelated domain-specific audit;
• reviewed exception.

Never delete historical audits/ reports during capability rollout.

Acceptance criteria

• Pace defines the desired-state representation for the canonical audit capability/version.
• A fresh fleet observation classifies current adoption and legacy variants.
• No-write planning distinguishes missing, current, drifted, incompatible, blocked, exempt, and migration states.
• Canonical Aether semantics are referenced/pinned rather than hand-copied into consumer repositories.
• Repository-specific configuration/extensions remain possible without overriding canonical contracts.
• At least three materially different canaries prove installation and discovery.
• Provider projections remain generated/managed and cannot silently diverge from the pinned Aether source.
• Existing historical audit reports are preserved.
• One bounded reviewable PR per repository is the default rollout unit.
• Repeated planning on unchanged inputs is deterministic.
• Rollout can remain advisory while infrastructure matures; missing audit capability does not automatically freeze unrelated development.

Related

• Fleet continuity rollout pattern: Roll out repository continuity handoffs across the Ego Hygiene fleet #26
• Repository Intelligence audits: egohygiene/relay#74
• Hygiene/conformance visibility: egohygiene/observatory#5

Non-goals

• Running a holistic audit of every repository in this issue.
• Automatically merging fleet PRs.
• Copying Aether source into every repository.
• Forcing every repository to use every Aether skill/agent.
• Making provider-specific prompt files canonical.