Per-repo skill for the Peat enterprise control plane — Rust binary/library with feature-gated CDC sinks, IDAM federation, and KMS integrations, plus a Svelte admin UI, Helm chart, Zarf manifest, and UDS bundle.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Per-repo skill for the Peat enterprise control plane — Rust binary/library with feature-gated CDC sinks, IDAM federation, and KMS integrations, plus a Svelte admin UI, Helm chart, Zarf manifest, and UDS bundle.
when_to_use
Editing files under peat-gateway/, reviewing peat-gateway PRs, debugging CDC / OIDC / KMS / multi-tenancy / formations API issues, working on the admin UI, or working on the Helm chart / Zarf packaging / UDS bundle.
verifies_with
cargo fmt --check, cargo clippy -- -D warnings, cargo test (default features), per-feature test/build for any changed feature, ui pnpm build for UI changes, helm template for chart changes.
peat-gateway SKILL
peat-gateway is the enterprise control plane for the Peat mesh. It provides multi-org tenancy, CDC (change-data-capture), IDAM federation (OIDC), envelope encryption (AWS KMS / Vault), and a formations API for managing peer / document / certificate state. The repo combines a Rust binary + library (peat_gateway), a Svelte admin UI in ui/, a Helm chart, a Zarf manifest, and a UDS bundle. CDC sinks (NATS, Kafka, webhook), identity providers (OIDC), storage backends (Postgres), and KMS integrations are all behind feature flags so deployments compile only what they need — and CI splits the test job by feature to avoid OOM on standard runners.
When this skill applies
Editing any file under src/ (control plane, formations API, CDC dispatcher, OIDC, KMS, tenant manager)
Editing ui/ (Svelte admin UI)
Editing chart/peat-gateway/ (Helm chart) or zarf.yaml (Zarf packaging) or bundle/uds-bundle.yaml (UDS bundle)
Touching the feature-flag matrix in Cargo.toml
Working on a feature-gated sink (NATS, Kafka, webhook), identity provider (OIDC), storage backend (Postgres), or KMS integration (AWS KMS, Vault)
Bumping the pinned peat-mesh version (currently =0.9.0-rc.1)
Scope
In scope:
Multi-org tenancy and tenant manager logic
CDC watcher state machine and event dispatcher
Sink implementations (behind feature flags)
IDAM federation (OIDC, RFC 7662 token introspection)
Envelope encryption integrations (AWS KMS, Vault)
Formations API endpoints (peer, document, certificate)
Admin UI in ui/ (Svelte + Vite + pnpm)
Helm chart, Zarf manifest, UDS bundle
Per-feature CI configuration (the OOM-prevention split)
Out of scope (route elsewhere):
Mesh transport / sync semantics → peat-mesh/SKILL.md
BLE transport → peat-btle/SKILL.md
OCI registry sync → peat-registry/SKILL.md
Top-level shared types/traits — consider whether the change belongs in peat/peat-protocol or peat/peat-schema
Production cluster operations / GitOps configs that consume this chart — separate ops repos
Workflow
Orient. Read peat/SKILL.md (ecosystem) if accessible. Read this file. Read CONTRIBUTING.md. Skim relevant docs/ (architecture, cdc, identity-federation, multi-tenancy, etc.) for the area you're touching. git status, git log -10.
Locate the spec. Confirm the task has a GitHub issue with Context / Scope / Acceptance / Constraints / Dependencies. If not, stop and ask the user.
Plan. Produce a 1–5 step plan. Cross-check against ecosystem hard invariants (transport agnosticism, dependency direction, async runtime is Tokio) and the scope guards below.
Implement. Branch from main per the trunk-based convention. Vertical slices, one concern per commit. Keep new sinks / providers / backends behind their own feature flags.
Verify. Run every command in the verification checklist below. Capture output.
Hand off. Open PR against main referencing the issue. Single concern per PR — squash-merge applies. After opening the PR, self-review the full diff against the originating issue/plan and post findings as a PR comment covering correctness gaps (missing error handling, edge cases, panics), security concerns (secrets in memory, missing zeroization, input validation), missing test coverage, and deviations from issue requirements. Fix actionable findings before declaring the PR ready — the comment serves as a review record.
Verification (exit criteria)
A session in this repo is not done until each of these produces evidence:
cargo fmt --check exits 0
cargo clippy -- -D warnings exits 0 (run with the relevant feature combo for changed code)
cargo test exits 0 (default features = nats, webhook, oidc)
If the change touches code under any non-default feature: cargo test --features <flag> for each affected feature (kafka, postgres, aws-kms, vault); CI splits these to avoid OOM, run them locally one at a time
If the change is broad enough to span features: cargo build --features full confirms the all-features build still links
If ui/ was touched: cd ui && pnpm install && pnpm build succeeds
If chart/ or zarf.yaml or bundle/uds-bundle.yaml was touched: helm template chart/peat-gateway renders cleanly; for Zarf/UDS bundle changes, run the corresponding packaging command
If a feature-gated sink / provider / backend interaction changed: the feature's optional runtime dependency (NATS server, Kafka broker, Postgres, Vault) is exercised via the integration tests, not just the build
"Seems right" or "the diff looks correct" is never sufficient.
Anti-rationalization
Excuse
Rebuttal
"This change is too small to need a test."
If it's worth changing, it's worth one assertion. Add the test.
"I'll fix the clippy warning later."
The CI gate is -D warnings. There is no later.
"Default features cover most cases — I don't need to test the kafka/postgres/vault paths."
Feature flags exist because deployments differ. Each feature combination is a real production target. Run cargo test --features <affected>.
"I'll add the new sink straight to the CDC dispatcher inline — it's just one place."
New sinks live behind their own feature flags so deployments stay lean and OOM-on-CI doesn't recur. Add the feature flag, gate the dep, gate the dispatcher branch.
"Splitting CI by feature is annoying — I'll combine the jobs."
The split exists because combined builds OOM on standard runners (efe2e8f). Don't recombine without a verified solution.
"I'll bump peat-mesh to the latest RC for the new feature."
The pin (=0.9.0-rc.1) is intentional. Bumps need full integration validation and possibly chart/Zarf updates.
"Admin UI changes don't need a build — they're just frontend."
The UI ships as static assets via the Helm chart. Build it (cd ui && pnpm build) so deploy-time bundling doesn't break.
"I'll add the new identity provider as a sibling to OIDC, no feature flag — it's just config."
OIDC is feature-gated for a reason. Identity providers carry transitive security/runtime deps; gate the new one.
"I'll inline the KMS call without zeroization — it's only in memory briefly."
KMS material lives in security-sensitive paths. Use the existing zeroization patterns; don't introduce un-zeroized secret handling.
"I'll suppress the new clippy warning by adding #[allow] everywhere."
Suppression hides the signal. Either fix the underlying issue or, if it's genuinely a false positive, gate with a single #[allow] and a comment explaining why.
Scope guards
Touch only files the issue/user asked you to touch.
Do not edit other peat-* repos. Cross-repo work goes in a separate PR in that repo, linked through a tracking issue.
Keep new sinks / providers / backends behind their own feature flags. Do not pull optional runtime deps (NATS, Kafka, Postgres, Vault clients) into the default feature set.
Do not bleed control-plane logic into mesh code (and vice versa). The seam is peat-mesh's public API (used here with automerge-backend and broker features).
Do not commit secrets, KMS material, or absolute paths in chart/, zarf.yaml, bundle/, or test fixtures.
Do not configure git to bypass GPG signing or use --no-verify to skip pre-commit hooks.
Gotchas
Add an entry each time a session produces output that needed correction. One line per gotcha plus a Why: line.
