fleet-bootstrap

fleet-bootstrap is the cross-repo remediation skill: it converges 6+ repos onto a standard Octopus configuration from one control file, running octopus setup in each. It closes the drift that audit-fleet detects.

Two realities of a real fleet shape it — and break the naïve “one canonical .octopus.yml” idea:

• Multi-stack — a .NET repo needs different bundles than a frontend repo, so the standard is layered, not monolithic.
• Legacy code — a hard guardrails flip breaks legacy commits instantly, so adoption is phased (a ratchet), not all-at-once.

For the full end-to-end flow with diagrams, see Fleet setup flow.

The layered standard

Each repo’s manifest is composed as baseline ∪ stack profile ∪ tier:

• Baseline — stack-agnostic, every repo (agents, workflow, the workspace: reference, baseline bundles + roles).
• Stack profile — adds the stack’s bundles, selected by auto-detection (*.csproj → dotnet, package.json + framework → node/frontend, pyproject.toml → python) or pinned per repo for legacy/monorepo.
• Adoption tier — T0 / T1 / T2 (below).

Adoption tiers — the legacy ratchet

Loop-level hooks only act on what the agent edits going forward; the pre-commit (git) and CI layers are what break legacy. Tiers select which layers turn on:

Tier	Turns on	Seeds	Legacy
T0	bundles/roles/rules; hooks: false	agent config, rules	zero enforcement — safe now
T1	hooks: true + enforce-ide	.editorconfig per stack	catches new drift; no commit/CI gate
T2	+ enforce-precommit + qualityWorkflow	.husky/pre-commit + CI	full blocking floor — when clean

Legacy repos start at T0/T1; the manager bumps tier: and re-runs.

One control file — fleet.yml

Lives in the workspace: repo. Carries the baseline, the profiles, the tiers, and the repos: map (each repo’s optional profile + tier). audit-fleet and fleet-bootstrap read the same repos: list.

Thin orchestrator

The skill writes only the .octopus.yml; the actual seeding — rules, hooks, agent config, .editorconfig, .husky — is delegated to octopus setup, exactly as for a single repo.

Safe by default

• Dry-run by default — preview the per-repo diff, write nothing.
• Per-key merge — converges baseline + tier, keeps profile-justified local additions, flags arbitrary keeps and any tier de-escalation (never silent removal); *.local.md overrides survive.
• --apply writes + runs setup; --yes for a trusted batch; --pr opens a guarded branch + PR per repo (never pushes to main); --from-audit scopes to the repos audit-fleet flagged.

The trade-offs follow the non-destructive merge policy and workspace config-template precedence.

Source: skills/fleet-bootstrap/SKILL.md