Skip to content

Fleet Go demand (pre-migration input)

What the fleet's Go repositories import today, and how much of it capabilities.yaml already claims to replace. This is the demand side of the pre-migration epic (#429): it says what golusoris must provide for each consumer to drop its third-party dependencies.

Scope is Go only. The fleet's TypeScript, Rust and Python dependencies are outside golusoris's remit and are excluded deliberately.

Method

Every go.mod in each repository is parsed for direct requirements, excluding vendor/, testdata/, node_modules/, _audit/, and the agent scratch directories .claude/worktrees/ and .workingdir/. Requirements resolving to another module in the same repository, or to another first-party fleet module, are not third-party demand.

Counted straight from go.mod rather than through praetorctl needs scan, because that scanner reads only the repository-root go.mod. It reports one dependency for a repo whose nested modules pull dozens, and fails with no matching language analyzer on any repo without a root module (goenvoy, k8s, watershed). Its numbers are not a usable migration input yet.

Summary

Metric Value
Consumer repositories with Go 9 of 10
Distinct third-party packages in demand 96
Already claimed by capabilities.yaml 38 (40%)
Gaps with no golusoris equivalent 58

Per repository

Repository Go modules Third-party Covered Coverage
VMAFx/vmafx 5 77 26 34%
lusoris/k8s 14 9 7 78%
lusoris/venio 1 8 7 88%
lusoris/20-watts-was-enough 16 8 4 50%
cordanaLLM/imago 2 4 3 75%
cordanaLLM/praetor 1 1 1 100%
lusoris/praetor 1 1 1 100%
golusoris/goenvoy 69 0 0 n/a
lusoris/watershed 3 0 0 n/a
cordanaLLM/Aegis-OS 0 0 0 n/a

Three entries need reading rather than scoring:

  • golusoris/goenvoy has 69 modules and zero third-party requirements. The metadata and arr clients are deliberately stdlib-only, so there is nothing to migrate.
  • cordanaLLM/Aegis-OS contains no Go at all (no go.mod, no .go files); it is a Rust workspace. golusoris has nothing to offer it, recorded here so the question is not asked again.
  • cordanaLLM/praetor and lusoris/praetor each have exactly one dependency, gopkg.in/yaml.v3, which core/codec/yaml now replaces. Both are one import rewrite from zero third-party dependencies, which is the point of the core/ carve-out.

Shared demand (two or more consumers)

Highest leverage: each of these removes a dependency from more than one repository.

Consumers Status Package Used by
3 covered github.com/modelcontextprotocol/go-sdk VMAFx/vmafx, cordanaLLM/imago, lusoris/k8s
3 covered modernc.org/sqlite VMAFx/vmafx, lusoris/20-watts-was-enough, lusoris/k8s
2 covered gopkg.in/yaml.v3 cordanaLLM/praetor, lusoris/praetor
2 covered github.com/prometheus/client_golang VMAFx/vmafx, lusoris/k8s
2 covered github.com/sirupsen/logrus VMAFx/vmafx, lusoris/20-watts-was-enough
2 covered github.com/spf13/cobra VMAFx/vmafx, cordanaLLM/imago
2 covered k8s.io/api VMAFx/vmafx, lusoris/k8s
2 covered k8s.io/apimachinery VMAFx/vmafx, lusoris/k8s
2 covered k8s.io/client-go VMAFx/vmafx, lusoris/k8s
2 gap github.com/stretchr/testify cordanaLLM/imago, lusoris/venio

Gaps (58)

Packages the fleet imports that capabilities.yaml does not claim. Each is either a capability golusoris could add, or a deliberate non-goal worth recording as one.

Consumers Package Used by
2 github.com/stretchr/testify cordanaLLM/imago, lusoris/venio
1 cloud.google.com/go/bigquery VMAFx/vmafx
1 cloud.google.com/go/pubsub VMAFx/vmafx
1 cloud.google.com/go/storage VMAFx/vmafx
1 contrib.go.opencensus.io/exporter/stackdriver VMAFx/vmafx
1 github.com/anchore/go-logger lusoris/20-watts-was-enough
1 github.com/anchore/stereoscope lusoris/20-watts-was-enough
1 github.com/anchore/syft lusoris/20-watts-was-enough
1 github.com/bombsimon/logrusr/v2 VMAFx/vmafx
1 github.com/bradleyfalzon/ghinstallation/v2 VMAFx/vmafx
1 github.com/c-bata/goptuna VMAFx/vmafx
1 github.com/containerd/nri lusoris/k8s
1 github.com/go-git/go-billy/v5 VMAFx/vmafx
1 github.com/go-git/go-git/v5 VMAFx/vmafx
1 github.com/gobwas/glob VMAFx/vmafx
1 github.com/golangci/golangci-lint/v2 VMAFx/vmafx
1 github.com/google/addlicense VMAFx/vmafx
1 github.com/google/go-cmp VMAFx/vmafx
1 github.com/google/go-containerregistry VMAFx/vmafx
1 github.com/google/go-github/v46 VMAFx/vmafx
1 github.com/google/go-github/v82 VMAFx/vmafx
1 github.com/google/ko VMAFx/vmafx
1 github.com/google/osv-scanner/v2 VMAFx/vmafx
1 github.com/google/uuid VMAFx/vmafx
1 github.com/goreleaser/goreleaser/v2 VMAFx/vmafx
1 github.com/grafeas/kritis VMAFx/vmafx
1 github.com/h2non/filetype VMAFx/vmafx
1 github.com/hmarr/codeowners VMAFx/vmafx
1 github.com/in-toto/attestation VMAFx/vmafx
1 github.com/in-toto/in-toto-golang lusoris/20-watts-was-enough
1 github.com/jszwec/csvutil VMAFx/vmafx
1 github.com/mcuadros/go-jsonschema-generator VMAFx/vmafx
1 github.com/microsoft/azure-devops-go-api/azuredevops/v7 VMAFx/vmafx
1 github.com/moby/buildkit VMAFx/vmafx
1 github.com/olekukonko/tablewriter VMAFx/vmafx
1 github.com/onsi/ginkgo/v2 VMAFx/vmafx
1 github.com/onsi/gomega VMAFx/vmafx
1 github.com/otiai10/copy VMAFx/vmafx
1 github.com/parquet-go/parquet-go VMAFx/vmafx
1 github.com/rhysd/actionlint VMAFx/vmafx
1 github.com/shurcooL/githubv4 VMAFx/vmafx
1 github.com/shurcooL/graphql VMAFx/vmafx
1 github.com/sigstore/cosign/v2 VMAFx/vmafx
1 github.com/xeipuuv/gojsonschema VMAFx/vmafx
1 gitlab.com/gitlab-org/api/client-go VMAFx/vmafx
1 go.opencensus.io VMAFx/vmafx
1 go.uber.org/fx VMAFx/vmafx
1 go.uber.org/mock VMAFx/vmafx
1 go.yaml.in/yaml/v2 VMAFx/vmafx
1 gocloud.dev VMAFx/vmafx
1 golang.org/x/net VMAFx/vmafx
1 golang.org/x/oauth2 VMAFx/vmafx
1 golang.org/x/sync VMAFx/vmafx
1 golang.org/x/sys lusoris/k8s
1 golang.org/x/text VMAFx/vmafx
1 google.golang.org/protobuf VMAFx/vmafx
1 mvdan.cc/sh/v3 VMAFx/vmafx
1 sigs.k8s.io/release-utils VMAFx/vmafx

Proposal

Synthesis of 58 classified fleet-demand gaps (Task N14) against the pre-migration epic (#429: #431 decoupling, #432 dependency substitution, #433 verification, #434 activation). Verified against the actual golusoris package tree and go.mod rather than taken at face value — several classified nearest_golusoris paths referenced a core/ prefix and packages (core/gitx, core/codec/yaml) that do not exist anywhere in the repository; those are corrected below.

Scale: 58 items, 51 VMAFx/vmafx-only, 7 shared with other fleet repos (lusoris/20-watts-was-enough ×4, lusoris/k8s ×2, cordanaLLM/imago + lusoris/venio ×1 via testify).

By cluster (consumers desc, effort asc within cluster)

Cluster Items Real target package Effort Unblocks
auth-crypto 7 new supplychain + auth/oauth2client L VMAFx, 20-watts-was-enough
cli-tui 5 clikit/tui (mostly non-goals) S VMAFx
cloud-provider-sdk 6 pubsub (GCP backend) + new container/registry M VMAFx
db-driver 1 new db/parquet M VMAFx
http-api 7 new integrations/github M VMAFx
infra-tooling 7 repo root (fx); rest mostly non-goal/new S VMAFx
k8s 3 k8s/operator (NRI) M lusoris/k8s
media-video 1 storage/safety S VMAFx
misc 9 archive + jsonschema S VMAFx, 20-watts-was-enough
ml-ai 1 new ai/optimize L VMAFx
observability 5 log (already covered) + supplychain (osv-scanner) S VMAFx, 20-watts-was-enough
testing 6 testutil/fxtest / testutil/pact S cordanaLLM/imago, lusoris/venio, VMAFx

Corrections found by re-reading the package tree

Four nearest_golusoris values in the classified data pointed at a core/ prefix that does not exist in the repo (find . -maxdepth 1 -type d -name core returns nothing): go.uber.org/fx → core, github.com/google/uuid → golusoris/golusoris/core/id (also malformed), github.com/anchore/go-logger / logrusr → core/log. The real paths are the repo root (fx), id/, and log/ respectively — all top-level packages, verified present with AGENTS.md + implementation files.

Two items claimed an existing wrapper that does not exist: go-git (claimed already used by core/gitx) and go.yaml.in/yaml/v2 (claimed aliased by core/codec/yaml on v3) — grep of the full tree found zero go-git references and go-git is absent from go.mod entirely; go.yaml.in/yaml v2 and v3 are listed only as indirect dependencies, confirming no direct-use wrapper exists. Both are gaps to build, not aliases to document.

No cross-chunk disagreements (the same package classified two different ways) were found in this 58-item set — every package appears once. The issues above are single-source path/fact errors, not conflicting classifications.

VMAFx-only demand (51 of 58 items, ~88% of this batch)

The overwhelming majority of this classified batch is VMAFx/vmafx-only demand: cloud-provider SDKs (BigQuery, GCS, Pub/Sub, Azure DevOps), the entire supply-chain-security cluster (cosign, in-toto, kritis, osv-scanner), CI/build tooling (ko, goreleaser, buildkit, golangci-lint, addlicense, actionlint), and most of http-api/cli-tui/testing. Given the fleet-wide context (77 of 96 third-party packages are VMAFx's), the maintainer should weigh how much of golusoris' roadmap should be steered by one consumer's dependency footprint versus the 7 items shared across lusoris/20-watts-was-enough, lusoris/k8s, cordanaLLM/imago, and lusoris/venio — the latter arguably deserve priority despite lower per-cluster item counts, since fixing them unblocks decoupling (#431) for repos beyond VMAFx.

First sprint (10 cheapest, highest fleet-unblock items)

  1. storage/safety: h2non/filetype magic-byte detection (S)
  2. archive: otiai10/copy recursive directory copy (S)
  3. testutil/pact: ginkgo fx-aware BDD lifecycle wrapper (S)
  4. testutil: csvutil-based fixture codec (S)
  5. jsonschema: go-jsonschema-generator for round-trip schema gen (M)
  6. pubsub: GCP Pub/Sub backend beside kafka/nats (M)
  7. k8s/operator: containerd/nri plugin hooks — 2 consumers (M)
  8. .needs.yaml: document ~30 non-goal packages with rationale — zero code, immediate readiness-score gain
  9. container/registry: new package wrapping go-containerregistry (M)
  10. Flag go-git / mvdan.cc/sh / go.yaml.in/yaml as unbuilt gaps (not aliases) in .needs.yaml before task #432 substitution runs, so the substitution step doesn't silently point at nonexistent packages

Non-goals declared (representative, full list per cluster above)

BigQuery, GCS, Azure DevOps API, gocloud.dev (cloud-provider-sdk — VMAFx's GCP/Azure-first architecture, not golusoris' S3/Postgres-shaped abstractions); anchore/stereoscope, anchore/syft, go-billy, gobwas/glob, hmarr/codeowners, x/text (stdlib-sufficient or too specialized); golangci-lint, addlicense, actionlint, ko, goreleaser, buildkit (standalone CI/CD tools, not libraries); gomega, testify, go.uber.org/mock (accepted standard testing deps); shurcooL/graphql, xeipuuv/gojsonschema, gitlab client-go (superseded by genqlient/santhosh-tekuri or provider-specific with no surface); ghinstallation (GitHub-App-specific auth, not a JWT primitive gap); protobuf, x/sys (transitive, no wrapping value).