diff --git a/.claude/rules/skill-guidance.md b/.claude/rules/skill-guidance.md
index 854366fda9..a754a06e5f 100644
--- a/.claude/rules/skill-guidance.md
+++ b/.claude/rules/skill-guidance.md
@@ -18,6 +18,8 @@ This project has guided skills for common workflows. **Proactively suggest the r
 | `/scaffold-snowflake-connector` | Add a new Snowflake-connector data source or integration                          |
 | `/packages-worker-setup` | First-time setup of packages-db and github-repos-enricher for a new engineer    |
 | `/packages-worker-add-entrypoint` | Scaffold a new sibling worker inside packages_worker (npm, OSV, scorecard, etc.) |
+| `/write-unit-tests` | Add or improve Vitest unit tests for business logic, DAL, or server modules |
+| `/write-api-e2e-tests` | Add or change Public API e2e / smoke / contract tests |
 
 ## Trigger Phrases
 
@@ -57,3 +59,14 @@ This project has guided skills for common workflows. **Proactively suggest the r
 - "Add a new packages worker", "scaffold a sibling worker", "new entry point in packages_worker"
 - "Add npm ingestion", "add OSV worker", "add scorecard runner"
 - Any request to create a new `src/bin/*.ts` worker inside `packages_worker`
+
+**`/write-unit-tests`** — match any of these intents:
+- "Write unit tests", "add a unit test", "cover this with Vitest"
+- "Test this function", "add DAL tests", "unit test for affiliations/merges/inference"
+- Any request for focused server/unit test coverage (not HTTP/API contract tests)
+- Not for `packages_worker` (excluded from `pnpm test:server` for now)
+
+**`/write-api-e2e-tests`** — match any of these intents:
+- "Write API e2e tests", "add smoke tests", "Public API contract tests"
+- "Cover this endpoint end-to-end", "API regression test"
+- Any request for HTTP/API e2e coverage of Public API behaviour
diff --git a/.claude/skills/write-api-e2e-tests/SKILL.md b/.claude/skills/write-api-e2e-tests/SKILL.md
new file mode 100644
index 0000000000..9d7e70ebe8
--- /dev/null
+++ b/.claude/skills/write-api-e2e-tests/SKILL.md
@@ -0,0 +1,66 @@
+---
+name: write-api-e2e-tests
+description: >
+  Write API end-to-end tests. Use when adding or changing API endpoints, or when
+  the user asks for API e2e, smoke, or contract tests.
+allowed-tools: Bash, Read, Glob, Grep, Edit, Write, AskUserQuestion
+---
+
+# Write API end-to-end tests
+
+Write or extend API end-to-end tests.
+
+## When to use
+
+- New or changed API endpoints.
+- User asks for API e2e, smoke, or contract tests.
+- Critical API behaviour needs regression coverage.
+
+## When not to use
+
+- Domain or SQL correctness → use `write-unit-tests`.
+- Temporal, OpenSearch, or other eventual side effects outside the documented API e2e scope.
+- Creating a new testing framework or suite style.
+
+## Source of truth
+
+Read before writing. Follow these ADRs and existing suite structure; do not
+invent a parallel testing style.
+
+- [ADR-0012](../../../docs/adr/0012-api-e2e-test-architecture.md) — runtime, isolation, supported surfaces, scope, and assertions.
+- [ADR-0013](../../../docs/adr/0013-api-e2e-test-suite-design.md) — suite organisation, helpers, and conventions.
+
+Current default entrypoint:
+
+- `.github/scripts/public-api-e2e-tests.sh`
+
+## Workflow
+
+1. Identify the API surface. Default to Public API unless the user specifies otherwise.
+2. Read ADR-0012 and ADR-0013.
+3. Add or extend the appropriate suite and register it if required.
+4. Run the affected suite locally and fix failures until green.
+5. If required fixtures cannot be created through the API, prefer testing supported scenarios and explain any coverage gaps instead of seeding the database directly.
+6. Suggest production testability improvements only when they make the API easier to test, and ask before changing production code.
+
+## Run
+
+Export the environment variables required by the suite entrypoint.
+
+```bash
+bash .github/scripts/public-api-e2e-tests.sh
+```
+
+Refer to ADR-0012 and the suite entrypoint for environment setup, reset behaviour,
+and local development workflows.
+
+## Guardrails
+
+- Keep tests focused on observable API behaviour.
+
+## Output
+
+- Suites and cases added
+- How to re-run
+- Coverage gaps, if any
+- Optional testability suggestions
diff --git a/.claude/skills/write-unit-tests/SKILL.md b/.claude/skills/write-unit-tests/SKILL.md
new file mode 100644
index 0000000000..aecdc54282
--- /dev/null
+++ b/.claude/skills/write-unit-tests/SKILL.md
@@ -0,0 +1,72 @@
+---
+name: write-unit-tests
+description: >
+  Write focused Vitest unit tests. Use when adding or improving unit tests for
+  business logic, data access, common services, or other server modules.
+allowed-tools: Bash, Read, Glob, Grep, Edit, Write, AskUserQuestion
+---
+
+# Write unit tests
+
+Write focused unit tests that follow the project's testing conventions.
+
+## When to use
+
+- User asks to add or improve unit tests.
+- A change touches high-blast-radius logic (affiliations, merges, identity resolution, timelines, inference).
+- A PR needs confidence in a pure or Postgres-backed function.
+
+## When not to use
+
+- Public or HTTP contract coverage → use `write-api-e2e-tests`.
+- `packages_worker` (excluded from `pnpm test:server` for now; own vitest/packages-db).
+- Temporal, Redis, or OpenSearch fixtures (not available yet).
+- Broad "increase coverage %" requests without a clear unit under test.
+
+## Source of truth
+
+Read before writing. Follow these ADRs; do not invent a parallel testing style.
+
+- [ADR-0008](../../../docs/adr/0008-how-we-write-unit-tests.md) — scenarios, `describe` grouping, assertions, mocking, and shared setup.
+- [ADR-0007](../../../docs/adr/0007-test-factory-primitives-and-defaults.md) — factories and defaults.
+
+## Workflow
+
+1. Identify the unit under test (one function or decision path). Colocate tests as `<file>.test.ts`.
+2. Read ADR-0007 and ADR-0008. Skim the nearest existing test in the same area if one exists.
+3. Compose fixtures using `@crowd/test-kit` (`withQx` for Postgres-backed tests; factories and opt-in defaults per ADR-0007).
+4. Write focused scenarios following ADR-0008 (grouping, naming, assertions, and mocking).
+5. Run the affected tests and fix failures until green.
+6. If production code is difficult to test, suggest a small testability seam and ask before changing production code.
+
+## Run
+
+Start the test database when needed:
+
+```bash
+./scripts/cli scaffold up-test
+```
+
+Run a focused test file:
+
+```bash
+pnpm test:server -- path/to/file.test.ts
+```
+
+Optional:
+
+```bash
+pnpm test:changed
+pnpm test:watch -- path/to/file.test.ts
+```
+
+## Guardrails
+
+- Prefer critical behaviours over trivial getters, setters, and thin wrappers.
+- Keep production behaviour unchanged unless the user explicitly asks for a testability improvement.
+
+## Output
+
+- Scenarios covered
+- How to re-run
+- Optional testability suggestions
diff --git a/docs/adr/0008-how-we-write-unit-tests.md b/docs/adr/0008-how-we-write-unit-tests.md
index 3e6f84f2fb..5de6402a73 100644
--- a/docs/adr/0008-how-we-write-unit-tests.md
+++ b/docs/adr/0008-how-we-write-unit-tests.md
@@ -25,6 +25,8 @@ reuse is real.
 - Setup should make that story obvious (concrete labels, dates, relationships).
 - Do not combine unrelated behaviors in a single test.
 - Prefer fewer tests that fail clearly over many that pass for the wrong reason.
+- When a file covers more than one function, group cases with
+  `describe('<functionName>', …)` and put named scenarios inside.
 
 ### Assertions
 
@@ -33,6 +35,18 @@ reuse is real.
 - Avoid loose probes (`some`, `greaterThan`, optional finds) unless the claim
   truly is approximate.
 
+### Mocking
+
+- Do not mock the data-access layer or the unit under test. Postgres-backed
+  tests use a real test database and factories; query behavior is part of what
+  we ship.
+- Reach for mocks only at external network boundaries where the real dependency
+  would hurt reliability or speed (outbound HTTP, LLM clients, Slack/email
+  senders, Auth0 JWKS in end-to-end tests).
+- Control non-determinism in-process when needed (e.g. fake timers for “now”).
+- If a test seems to need a mock outside that list, the unit is often too large:
+  extract the pure part, unit-test it, and let the outer path stay real.
+
 ### What to share
 
 | Layer | Where | Belongs |